5)spectroscopic methods

5 Spectroscopic Methods

� 2008 by Taylor & Fr

for Texture and StructureAnalyses

Pilar Barreiro Elorza, Natalia Hernández Sánchez,Renfu Lu, Jesús Ruiz-Cabello Osuna,and David G. Stevenson

CONTENTS

5.1 Brief Overview on Approaches for Nondestructive Sensingof Food Texture ......................................................................................... 3785.1.1 Introduction on Food Texture ......................................................... 3785.1.2 Fruit and Vegetable Cell Walls in Relation to Texture .................. 3805.1.3 Spectroscopic Techniques to Measure Texture

of Starch-Based Cereal Foods......................................................... 3815.1.4 Spectroscopy for Fruit Texture and Structure Determination......... 3825.1.5 Vegetable Texture Measured by Spectroscopy............................... 3845.1.6 Texture of Dairy Products Measured by Spectroscopy .................. 3855.1.7 Spectroscopy for Measuring Texture of Nondairy

Processed Foods .............................................................................. 3865.1.8 Summary ......................................................................................... 386

References ............................................................................................................. 3875.2 Spectroscopic Technique for Measuring the Texture

of Horticultural Products: Spatially Resolved Approach .......................... 3915.2.1 Introduction ..................................................................................... 3915.2.2 Light Propagation in Scattering-Dominant Biological Materials ... 393

5.2.2.1 Scattering and Absorption ................................................. 3935.2.2.2 Diffusion Theory Model .................................................... 3955.2.2.3 Steady-State Solutions ....................................................... 396

5.2.3 Hyperspectral Imaging Technique for Measuringthe Optical Properties of Horticultural Products............................. 3985.2.3.1 Principle and Instrumentation............................................ 3995.2.3.2 Procedures of Determining the Optical Properties ............ 402

5.2.4 Applications..................................................................................... 4055.2.4.1 Optical Properties of Fruits and Vegetables...................... 4055.2.4.2 Evaluation of Apple Fruit Firmness .................................. 409

ancis Group, LLC.

5.2.4.3 Estimation of Light Penetration Depths in Fruit ............... 4095.2.4.4 Monte Carlo Simulation of Light Propagation

in Apple Fruit .................................................................... 4115.2.5 Light-Scattering Technique Feasible for Assessing Fruit

Firmness in Practice ........................................................................ 4135.2.5.1 Wavelengths Selection....................................................... 4145.2.5.2 Instrumentation .................................................................. 4145.2.5.3 Mathematical Description of Light-Scattering Profiles ..... 4165.2.5.4 Fruit Firmness Assessment ................................................ 418

5.2.6 Conclusions and Needs for Future Research .................................. 419Acknowledgment .................................................................................................. 420References ............................................................................................................. 4215.3 NMR for Internal Quality Evaluation

in Horticultural Products............................................................................ 4235.3.1 Overview on Applications in Fruits and Vegetables ...................... 4245.3.2 Basics of NMR Relaxometry and NMR Spectroscopy .................. 426

5.3.2.1 Magnetic Moment of Nucleus and Its Excitation ............. 4265.3.2.2 Relaxation of Nucleus after Excitation.............................. 4295.3.2.3 Signal Detection during Relaxation................................... 4325.3.2.4 NMR Relaxometry and NMR Spectroscopy..................... 432

5.3.3 MRI Fundamentals .......................................................................... 4355.3.3.1 Image Acquisition and Reconstruction.............................. 4365.3.3.2 Effect of Movement on Image Quality ............................. 4385.3.3.3 Sequence Parameters and Their Effects

on MR Image Quality........................................................ 4405.3.3.4 Fast and Ultrafast MRI Sequences .................................... 441

5.3.4 Detailed View of Applications in Fruits ......................................... 4455.3.4.1 Maturity in Avocados ........................................................ 4455.3.4.2 Pit in Cherries and Olives.................................................. 4475.3.4.3 Internal Browning in Apples ............................................. 4485.3.4.4 Mealiness in Apples and Wooliness in Peaches ............... 4525.3.4.5 Internal Breakdown in Pears ............................................. 4535.3.4.6 Freeze Injury in Citrus....................................................... 4585.3.4.7 Seed Identification in Citrus .............................................. 460

5.3.5 Concluding Remarks ....................................................................... 462References ............................................................................................................. 464

5.1 BRIEF OVERVIEW ON APPROACHES FOR NONDESTRUCTIVESENSING OF FOOD TEXTURE

DAVID G. STEVENSON

5.1.1 INTRODUCTION ON FOOD TEXTURE

Food texture is an often underestimated quality trait that determines consumersatisfaction and the likelihood for the food product. One bad incidence of consumers

� 2008 by Taylor & Francis Group, LLC.

experiencing texture of foods that fail to meet their expectations can curtail futurepurchases of that food. Each type of food has specific textural attributes consideredideal, which frequently vary for consumers residing in different geographical loca-tions. While the sensations experienced during mastication of food define thetexture properties of food, visual cues and mechanical properties during handlingprovide the mind with information about the perceived texture before the food everenters the mouth. Visually, the shape, size, color, and incidence of cell structure orair spaces preempt our expectation of food texture. Touching or cutting the foodinstantly provides information on the likely firmness, plus depending on response tothis mechanical force, brittleness, adhesiveness, cohesiveness, elasticity, and springi-ness of food may be gauged. Slicing into food may also provide sounds providingfurther cues about the food texture such as crunchiness. Once food enters the mouth,initial touch attempts to deform food and provide some information about firmness.Slight shear from tongue movements provides stimuli reflecting the springiness,viscosity, and adhesiveness. The first few chews provide information on hardnessand brittleness, and subsequent chews mix saliva with the food, forming a cohesivebolus that experiences higher shear allowing advanced detection of brittleness,moistness, crispness, graininess, smoothness, creaminess, and adhesiveness.

The concepts of food texture are one of the easiest fields of science to compre-hend, but may be one of the toughest to obtain meaningful results. Since food textureis determined by our sensory perception, it is obvious to have panelists judging thetextural attributes to optimize measurements. However, it is challenging to obtainaccurate texture measurements from a sensory panel, because every individualperceives texture differently and it is difficult to exclude bias. Frequent problemsencountered include the following: (1) training of sensory panelists can be timeconsuming and expensive; (2) insufficient number of participants to enable accuratemeasurements of the true perceived textural properties of the total population; (3) toget sufficient panel size, participants are often recruited from within institutionconducting sensory panel and therefore have bias because of having too muchprior knowledge about the sensory panel objectives; (4) food products such as fruitsand vegetables tend to be seasonal and therefore difficult to study storage effects aspanelists know when produce is typically harvested; and (5) often difficult to providepanelists with control samples during storage studies to ensure panelists are consist-ent with their rating over time.

In an effort to save time and potentially obtain textural analysis from a greaternumber of food samples, instrumentation that attempts to simulate the sensoryperception of panelists was developed. Currently, the primary instruments used tomeasure food texture are the Instron universal testing machine (TAXT2 textureanalyzer) and Zwicki (1120, Zwick Materialpruefung Co. Ltd.). These instrumentsallow for food to undergo both compression or tensile stress, with potential to varythe shape and size of the probe contacting food, speed probe travels on contact withfood, the percentage compression or strain, and the number of times the probecontacts the food. The meaning of each food textural term is subjective, but to assistdevelopment of instrumental texture analysis, standardized definitions were estab-lished (Szcze�sniak 1963) and a procedure involving double compression of the foodwas developed (Freidman et al. 1963). Correlation between instrumental analysis of


food texture and sensory panel perception of texture has been well establis hed(Szcze�s niak 1987), but inst rumentat ion still has the disad vantag e of being a destruc-tive test.

Ideally, distrib utors of fresh produce and food manuf acturers want a rapid testthat can nonde structively evalua te the textural quality of food so that non-acc eptablefood products can be remo ved from mark et, avoiding food produce rs from having atarnished reput ation. The need for rapid, nondestruct ive evalua tion of foodtexture has resulted in the develo pment o f spect roscop ic met hods such as spati allyresolved spect roscop ic methods (see Secti on 5.2) and nuclea r magne tic resona nce(NMR) spectroscopy and its associated magnetic resonance imaging (MRI)(see Secti on 5.3).

5.1.2 FRUIT AND VEGETABLE CELL WALLS IN RELATION TO TEXTURE

Plant material consists of epidermis, vascular bundles, parenchyma, and thick-walledsupporting collenchyma and sclerenchyma cells. Parenchyma tissue is mainly usedfor food consumption because of its acceptable texture. Size and shape of cells,cytoplasm to vacuole ratio, intercellular space volume, cell-wall thickness, osmoticpressure, and solutes present all influence texture of fruit and vegetable parenchyma(Ilker and Szcze�sniak 1990). Adjacent cells are cemented together by the middlelamella, which when stronger than the cell walls, release liquid contents whencompressed. Pectin provides mechanical strength for the cell walls and adhesionbetween cells. Solid-state NMR has been demonstrated to effectively measurepolysaccharide mobility in plant cell walls (Jarvis and Apperley 1990).

Degradation of pectin resulting in softening of fruit and vegetable can bedetected by decreases in infrared band and magnetic resonance signals specific topectic substances. Pectin has characteristic infrared spectra with bands at 1101, 1026,and 957 cm�1 from uronic acids and absorption band at 1066 cm�1 from sugars.Antisymmetrical stretching of glycosidic bonds is observed at 1154 cm�1. The main13C-NMR signals from tomato pericarp were attributed to galacturonic acid in pectin,its methyl-ester substituent as well as the arabinose and galactose side chains, andrhamnose and acetyl esters. It was demonstrated from 13C-NMR spectra thatdifferent tomato varieties were enriched with loosely interacting pectic galactan ormethyl-ester side chains, which are important in cell-wall rigidity and adhesion indetermining texture. Uronic acids and protein peaks from near-infrared (NIR)spectra of tomato fruit were highly correlated with mealiness (negatively) andjuiciness (positively). NMR spectra detected differences in pectic arabinan sidechains, which have been linked to cell adhesion defects and perceived mealiness.

In a study of cell walls of lemons and maize, percentage crystallinity and crystalsize of cellulose microfibrils influence NMR and wide-angle x-ray scattering(WAXS), with higher crystallinity and larger crystal size having improved texturalproperties (Rondeau-Mouro et al. 2003). Smaller angles of reflection using WAXSindicate regions of food that are shifting from crystalline to amorphous. While plantcells have complex organization, there is some repeatability in arrangement ofmolecules within cells. This highly ordered structure has made whole fruit andvegetable tissues attractive food products to study the potential of spectroscopic


techniques to nondestructively elucidate their textural attributes. Presently, lessattention has been given to study implementation of spatially resolved spectroscopictechniques to determine texture of processed foods because many molecules aremodified during processing, water is redistributed, and majority of moleculesare intertwined resulting in a considerably greater heterogeneous material to studythat has greater problems with absorption and scattering of signals.

5.1.3 SPECTROSCOPIC TECHNIQUES TO MEASURE TEXTURE OF STARCH-BASED

CEREAL FOODS

Many studies have utilized NMR, MRI, and NIR to study water holding capacity instarch-based foods, primarily of cereal origin. Starch is composed of essentiallylinear amylose and highly branched amylopectin. MRI can experience problemswith cereal products owing to reduced transverse relaxation times. Proton NMRspectroscopy found that hardness of cooked rice was strongly correlated with spin–spin relaxation constants of protons (Ruan et al. 1997). Increasing starch concentra-tion results in decreasing T2 values in gels, reflecting the shorter distance watermolecules have on average to diffuse before interacting with starch, which conse-quently enhances proton decay because of decrease in water mobility, protonexchange, and cross-relaxation mechanisms (Hills et al. 1990; Yakubu et al. 1993).

In other studies involving rice, NIR spectra obtained in optical geometry fortransmittance readings could be successfully correlated to apparent amylose content,an important influence of cooked rice texture. White rice provided stronger correl-ations than brown rice probably because of bran in the latter (Villareal et al. 1994).Rice grains that were immature, slender, damaged, or dark had higher variance in thespectral intensities in the NIR as well as visible wavelength range.

Near-infrared reflectance spectroscopy (NIRS) data compared with textural attri-butes of cooked white rice, rated by sensory panelists, were able to correlate adhe-siveness, hardness, cohesiveness of mass (three chews), and toothpack, whilecohesiveness of mass (eight chews), roughness of mass, and toothpull were not wellcorrelated (Meullenet et al. 2002). NIRS of 62 Chinese rice flour samples successfullypredicted the paste viscosity parameters of setback and breakdown, but texturalparameters of chewiness, hardness, and gumminess were less related (Bao et al.2001). Texture attributes of 77 different short-, medium-, and long-grain rice cultivarshave been correlated with NIRS data. Hardness, initial starchy coating, cohesiveness ofmass, slickness, and stickiness as rated by sensory panelists were predicted by NIRspectral data. Important wavelengths contributing to the prediction models wereamylose, protein, and lipid contents. Several wavelengths contributing to models forstickiness contributed to model for amylose content, whereas several wavelengthscontributing to slickness model contributed to the protein model. Several wavelengthscontributing to models of amylose, protein, and lipid correlations contributed tohardness model. Wavelengths at regions 1654–1666 nm and 1966–1996 nm contrib-uted to initial starchy coating, cohesiveness of mass, and stickiness.

Strong positive correlations between cooked rice cohesiveness or adhesivenessand NIR measurements are because of observed absorbance of O��H and C��Howing to moisture (958 nm) and starch (878 and 979 nm), respectively. Negative


correlations betweenNIR spectra and hardness, springiness, gumminess, and chewinessare owing to absorbance of N��H of proteins (1018 nm) and C��H of oil (1212 nm).

Multiphase behavior in water relaxation studied by NMR has been observed inother starchy food systems such as flour dough (Leung et al. 1979) and bread (Chenet al. 1997; Engelsen et al. 2001). Amylose retrogradation contributes to increasedfirmness of foods and occurs so rapidly during cooling that often NMR analysiscannot be accomplished (Choi and Kerr 2003).

NMR has also been used to determine the texture of pasta. Stronger NMR signalintensity in the center of cooked lasagna sheets was shown to correspond with softertexture and water migration influx, while lower NMR signal intensity at exteriorreflected regions where water migrated from and had stiff texture (González et al.2000). Spin–spin relaxation time ofwater protons fromMRIwas used to detectmoisturedistribution in cooked spaghetti, and detected pasta with soft or brittle texture owing tomoisture homogeneity (Irie et al. 2004). Transverse relaxation time T2 images and mapsofwhite salted noodles demonstrated differences inwater status among noodles (Lai andHwang 2004). Increased cooking time resulted in decreased T2 values and softernoodles.

NIR spectra of wheat bread were highly correlated to firmness measured bytexture analyzer (TA) (Xie et al. 2003). Less variation in data was obtained fromNIR spectra among different bread batches compared with TA; therefore, NIRS wasmore precise at measuring both physical and chemical changes in bread staling, withextra advantage of being nondestructive. Physical changes are due to alterations inscattering properties as crystallinity develops during bread aging. NIRS also measureschemical modification during bread staling such as water loss and starch structuralchanges, and this is the most likely reason NIRS has superior detection results of breadstaling. Moisture loss and changes in starch crystallinity will affect firmness of bread.

Carbon-13 cross-polarization magic-angle-spinning nuclear magnetic resonance(13C-CP-MAS-NMR) spectra of bread crumb and crust found narrowing of anomericcarbon atom (C-1) peak, which also experiences displacement to a larger chemical shift,and loss of triplet characteristics could detect differences between rusk and crispytextures. The observed peak is most likely due to amylose–lipid complexes, but amorph-ous amylopectin and amylose could also contribute (Primo-Martín et al. 2007).

5.1.4 SPECTROSCOPY FOR FRUIT TEXTURE AND STRUCTURE DETERMINATION

A high proportion of the studies utilizing spectroscopic techniques for the determin-ation of texture has focused on fruit, especially apples. Watercore, distinctive topipfruit, results in reduced firmness because of the fruit while still on the treeexperiences saturated fluid in intercellular airspaces adjacent to vascular strands.Multi-slice MRI of fruit was shown to successfully locate regions that had highersignal intensity, reflecting where watercore was prevalent (Wang et al. 1988). Longerspin–spin T2 relaxation times and marginally shorter spin–lattice T1 relaxation timeshave been reported for watercore apples relative to unaffected apples (Clark andRichardson 1999). Quantification of spatial variation of gray tones from MRI hasbeen used to detect apples that varied in hardness, elasticity, and incidence of bruises(Létal et al. 2003).


Mealy apples have unplea sant texture in which cell s have becom e separatedwithout ruptu ring and hav e weak resi stance to compr ession from biting. Cells arepushed to the side rathe r than damag ed and texture ap pears as soft and dry. Time-resolved laser remittan ce spect roscop y (TRS) has been used to detect mealines s ofapples by utilizing visible and NIR laser s (com pare Se ction 1.3.2). Predic tion ofmealy apples was found to be more accurate using absorp tion and scattering TRScoef ficients than the textural param eters of soft or dry (Valéro et al. 2004 ). Using thetime-resolv ed approac h also the effective pathl ength can be calculated . In a sensorfusion approac h of time-re solved and continuous wave spect roscopy the Lambe rt–Beer law was used directly for ana lysing frui t and vegetables quality (Z ude et al.2008). Nondestr uctively record ed VIS-NIR spectra correl ated wel l wi th roughne ss,crunchiness , and meal iness of apples (Mehinagi c et al. 2003). Apple cultivars withdifferent texture were discr iminat ed by detect ion of two large peaks in NIR range at1440 and 19 40 nm, which most like ly corres pond to wat er. Rou ghness o f apples washighly correl ated to a coef ficient at wave length s 1886 a nd 2050 nm, whi ch mostlikely corres ponds to star ch and protein contents, respec tive ly. Crunchi ness of ap pleswas correl ated positive ly, while meal iness had a negative relat ionship wi th spectro-scopic data in the 680 –710 nm range and at 980 nm, which c orresponds to chloro-phyll absorb ance ba nds as well as starc h and water conten ts, respec tively.

Apple cult ivar diff erences were observ ed for correl ation betw een compr essiveproperties and laser scatter parameter s (C ho and Han 1999). Bio-yi eld and ruptureforces correl ated well with laser light mig ration in the tissue. Red laser c onsistentlyperformed better than green for d etermina tion of apple fi rmness, which may beexplained by laser beam absorption ab ility penetra ting b elow apple skin. Predic tionof firmne ss and, after wavelengt h selection (Qing et al. 2007a), simult aneously, offirmness and soluble solid content (Qing et al. 2007b) is possible. Laser light opticalpower is not important for determining apple firmness, as higher power lasers did notreduce erroneous results. Hyperspectral scattering readings provide useful spectraland spatial images of apple fruit that enables successful prediction of firmness (Lu2007) (see Section 5.2).

Internal quality defects of tropical fruit crops (Jagannathan et al. 1994) as well aspipfruit, stone fruit, and citrus (Chen et al. 1989) have been detected utilizing MRI.Rupturing of cells by freezing, which can occur in interior but hidden by externallayers, results in fruit softening and detection has been demonstrated in blueberries,kiwifruit, zucchini, and apples by measuring the reduction in magnetic susceptibilitybecause of changes in T2 relaxation from loss of cellular integrity (Duce et al. 1992;Gamble 1994; McCarthy et al. 1995; Kerr et al. 1997). Chilling injury, a tissuebreakdown disorder that occurs at above freezing temperatures in some subtropicalfruits, was not detected during cold storage for persimmons using MRI, but wasobserved when left at room temperature (Clark and Forbes 1994).

Peach woolliness is an undesirable texture trait similar to apple mealiness withsoft flesh that has absence of crispness and juiciness. Peach woolliness has beenstudied using MRI with less skewed MRI spectra T2 maps found for fruits withwoolliness. NIRS in conjunction with nondestructive impact tests was able to alsoidentify peaches exhibiting woolliness (Ortíz et al. 2001). Controlled atmosphericstorage can result or prevent undesirable fruit textural changes. MRI has been


demonstra ted to be effective in detect ing tissue breakdo wn in cores of pears(Wang and Wang 1989) stored at low oxygen, and illustrate p revention of woollybreakdo wn in nectar ines stored at low oxygen and elevated carbon dioxi de(Sônego et al. 1 995).

Water-soak ing is a physi ological disorder of mel ons charact erized by a glassytexture that can be d etected in fruits using NM R ima ging. Use of Fourier trans forminfrared (FT IR) diffuse re fle ctance spectroscop y (Fu et al. 2007) and time-domai ndiffuse re flectanc e spect roscop y (see Sectio n 1.3.2) can both effect ively predictfirmness of kiwifruit noninvasively.

5.1.5 VEGETABLE TEXTURE MEASURED BY SPECTROSCOPY

Likewise for fruit texture, spectroscopy techniques have been investigated for theirpotential to measure vegetable texture, especially potatoes. MRI has been shown todifferentiate raw potatoes that varied in some textural attributes when cooked. Three-quarters of variation in hardness and 50%–54% of adhesiveness and moistness ofcooked potatoes were predicted by MRI of raw potatoes, but mealy and grainytextures were poorly correlated with MR images. Prediction of mealiness was betterfrom vertical MR images rather than from entire or horizontal tuber regions (Thyboet al. 2000, 2004). Relaxation curves obtained from low-field 1H-NMR showed goodcorrelation with texture of cooked potatoes because NMR detected water content thatwas inversely related to starch content. Starch content was in fact the predominantdeterminant of cooked potato texture. Adhesiveness and springiness of cookedpotatoes were better predicted by NMR than compositional analysis (Thygesenet al. 2001), but later studies modeling NMR relaxation data of cooked potatoesfound good correlations with hardness, cohesiveness, adhesiveness, mealiness,graininess, and moistness (Povlsen et al. 2003). NIR spectra predicted well themoistness, waxiness, firmness, and mealiness of steamed potatoes adjudged bysensory panelists (Boeriu et al. 1998). Although O��H, N��H, and C��H stretchingovertones attributed to polysaccharides and proteins can be traced in spectrum, watercontributed the greatest to NIR spectra of potatoes. Another study found NIR spectraof raw potatoes were related to textural attributes (van Dijk et al. 2002). NIR spectrawere able to measure dry matter and starch content of potatoes, both importantcontributors to cooked texture. Correlations were established between NIR spectraldata and moistness, mealiness, crumbliness, waxiness, graininess, ability to mash,and firmness of cooked potatoes.

NIR spectra from cooked carrots were influenced by rate of water absorption andabsorption of carotenoids in visible light region (De Belie et al. 2003; Zude et al.2007). The phloem region of carrots provided better NIR spectra for determiningtexture compared with xylem tissues, possibly because phloem has larger volume,and higher sugar and carotenoid content. NMR relaxation and MRI can measure theabundance and spatial distribution of free and bound water. NIR spectra correlatedwell with cooked carrot hardness, crispness, and juiciness. Water is one of thestrongest absorbers of infrared, with three broad peaks in the NIR spectra observedfor fruits and vegetables with high water content. The peaks are at 970, 1450,and 1940 nm corresponding to the stretch of O��H molecular bonds of the second


and first overtones and O��H deformation, respectively. For FTIR, water prominentbands are at 3360 cm�1 for H��O stretching, 2130 cm�1 for water association, and1640 cm�1 for H��O��H bending vibration (Šáfá�r et al. 1994). A peak in visibleregion (450 nm) is due to absorption of carotenoids that decreased during cooking.The magnitude of reflectance change signifies differences in texture and relatedoptical scattering properties of cooked carrots.

5.1.6 TEXTURE OF DAIRY PRODUCTS MEASURED BY SPECTROSCOPY

Unlike fruit, vegetable, and meat texture, many dairy products have been processedand consist of a heterogeneous medium, creating additional challenges for relatingspectroscopic techniques to texture. Cheese texture, in particular, can be difficult tomeasure using traditional methods. Two-dimensional spin warp MRI was shown toeffectively detect water and liquid-lipid portions of cheddar, brie, and Danish bluecheeses during ripening by changes in relaxation time (Duce et al. 1995). Cavities incheese, such as fromage frais, could be detected due to their lack of response usingthree-dimensional missing pulse steady-state free precession NMR and Dixon chem-ical shift resolved imaging sequences.

Another method to nondestructively measure cheese texture is to use spectrafrom tryptophan fluorescence that is able to predict surface state, dry to watery ratio,texture length, and pastiness of soft cheeses by detecting the organization of proteinnetworks (Dufour et al. 2001). Different textures (firmness, disintegration, pastiness,graininess, springiness, crumbliness, and oval holes) of soft cheeses owing todifferences in molecular organization have also been identified using surface fluor-escence spectra. Semi-hard cheese texture during ripening was measured usingfluorescence and infrared spectra that detected differences in protein network(Mazerolles et al. 2001). NIR spectra have been used to determine springiness,pastiness, coherence, and hardness of semi-hard cheeses (Sørensen and Jepsen1998). Fluorescence spectra of Salers cheese were also found to have good correl-ation with firmness, adhesiveness, and springiness (Lebecque et al. 2001). NIRspectra of 20 Emmental cheeses correlated well with firmness and adhesiveness(Karoui et al. 2003). Crumbly texture of cheddar cheeses has been correlated wellwith NIR spectra (Downey et al. 2005). Models developed based on NIR spectrafrom processed cheeses were successful in predicting chewy, melting, creamy,fragmentable, firmness, rubbery, greasy, and mouthcoating textural attributesadjudged by sensory panelists. NMR can noninvasively determine microstructureof Grana Padano cheese (De Angelis Curtis et al. 2000).

Viscosity of milk can be measured using MRI. Signal-to-noise ratio (SNR) ofchocolate milk is decreased and velocity profile of flow is blurred compared withstrawberry milk because chocolate milk contains solid cocoa particles, whereasmagnetic resonance signal originates from aqueous protons (McCarthy et al. 2006).

Mid-infrared reflectance (MIR) spectrum represents absorption of all chemicalbonds having infrared activity between 4000 and 400 cm�1. Phospholipid-phasetransition to a firmer gel texture can be detected by MIR, as increasing temperatureresults in shift of bands associated with C��H and carbonyl stretching (Casaland Mantsch 1984). FTIR in conjunction with attenuated reflectance circumvents


sampling problems allowing investigation of textural changes resulting from aggre-gation and gelation of b-lactoglobulin (Dufour et al. 1998). Fluorescence spectros-copy can provide information on development of milk coagulation because offluorescence of vitamin A located in the fat globule–protein interactions (Dufourand Riaublanc 1997).

5.1.7 SPECTROSCOPY FOR MEASURING TEXTURE OF NONDAIRY

PROCESSED FOODS

Spectroscopy techniques have been less explored for measuring texture of processedfoods because of interweaving of molecules creating a less homogeneous medium.Cell wall softening measured by spatially resolved spectroscopy, which has beendiscussed previously, occurs after food is heated, frozen and thawed, brined, or air-dried. Pectin solubilization occurs in acidic and alkaline conditions, and demethoxy-lation occurs during heating. Carboxyl groups can form salt linkages resulting inincreased firmness.

French fries are desirable when texture is crisp and moist rather than soggy.Time-domain NMR is a rapid and accurate nondestructive technique that canmeasure the nuclear spin–spin T2 relaxation times enabling detection of water andoil mobility and location within French fries and development of optimum fryingconditions that produce desirable texture (Hickey et al. 2006).

Three-dimensional MR images of chocolate confectioneries can completelyresolve the internal and external structure, thereby providing potential tool to assesstexture of this heterogeneous food (Miquel et al. 1998).

Static light scattering can be utilized to determine droplet size in oil in wateremulsions with starch or gums added as thickeners. Droplet size influences texture,and polysaccharides added stabilize emulsion against creaming by enhancing con-tinuous phase viscosity owing to gel network formation (Quintana et al. 2002).Decrease in average fat globule size reduces creaming velocity and increases emul-sion stability (Desrumaux and Marcand 2002). Light scattering detects changes in oilglobule size owing to Brownian motion and probability of collision and coalescenceboth increase during processing that alters texture.

Small-angle x-ray scattering (SAXS) can detect textural changes because ofagglomeration of protein and carrageenan. Lower intensity indicates small differencesin electronic intensity between scattered particles and their surrounding (Mleko et al.1997). T2 parameter of starch-based heated sauces measured by NMR relaxometrywas found to be a good indicator of viscosity (Thebaudin et al. 1998).

5.1.8 SUMMARY

Measuring food texture accurately is very challenging and conventionally has beenachieved by either sensory panels or instrumentation. Food industry’s desire for rapidnondestructive evaluation of food products has led to the emergence of noninvasivetexture analysis utilizing spatially resolved spectroscopy such as NIR, FTIR, NMR,and MRI. Significant advancements have already been made in establishing rela-tionships between these nondestructive tests and texture determined by sensory


panelists and instruments, thereby leading the way for a future in which spatiallyresolved spectroscopic techniques are standard quality-control procedures installedon conveyor belts of fresh produce and processed foods to ensure desired texture isdelivered to consumers without defects.

REFERENCES

Bao, J.S., Y.Z. Cai, and H. Corke. 2001. Prediction of rice starch quality parameters by near-infrared reflectance spectroscopy. Journal of Food Science 66: 936–939.

Boeriu, C.G., D. Yüksel, R. van der Vuurst De Vries, T. Stolle-Smits, and C. van Dijk. 1998.Correlation between near infrared spectra and texture profiling of steam cooked pota-toes. Journal of Near Infrared Spectroscopy 6: A291–A297.

Casal, H.L. and H.H. Mantsch. 1984. Polymorphic phase behavior of phospholipidmembranes studied by infrared spectroscopy. Biochimica et Biophysica Acta 779:382–401.

Chen, P., M.J. McCarthy, and R. Kauten. 1989. NMR for internal quality evaluation of fruitsand vegetables. Transactions of the American Society of Agricultural Engineers 32:1747–1753.

Chen, P.L., Z. Long, R. Ruan, and T.P. Labuza. 1997. Nuclear magnetic resonance studiesof water mobility in bread during storage. Lebensmittel-Wissenschaft und -Technologie30: 178–183.

Cho, Y.-J. and Y.J. Han. 1999. Nondestructive characterization of apple firmness by quantita-tion of laser scatter. Journal of Texture Studies 30: 625–638.

Choi, S.G. and W.L. Kerr. 2003. Water mobility and textural properties of native andhydroxypropylated wheat starch gels. Carbohydrate Polymers 51: 1–8.

Clark, C.J. and S.K. Forbes. 1994. Nuclear magnetic resonance imaging of the development ofchilling injury in Fuyu persimmon (Diospyros kaki). New Zealand Journal of Crop andHorticultural Science 22: 209–215.

Clark, C.J. and C.A. Richardson. 1999. Observation of watercore dissipation in ‘Braeburn’apple by magnetic resonance imaging. New Zealand Journal of Crop and HorticulturalScience 27: 47–52.

De Angelis Curtis, S., R. Curini, M. Delfini, E. Brosio, F. D’Ascenzo, and B. Bocca. 2000.Amino acid profile in ripening of Grana Padano cheese: An NMR study. Food Chem-istry 71: 495–502.

De Belie, N., D.K. Pedersen, M. Martens, R. Bro, L. Munck, and J. De Baerdemaker. 2003.The use of visible and near-infrared reflectance measurements to assess sensory changesin carrot texture and sweetness during heat treatment. Biosystems Engineering 85:213–225.

Desrumaux, A. and J. Marcand. 2002. Formation of sunflower oil emulsions stabilized bywhey proteins with high-pressure homogenization (up to 350 mPa): Effect of pressureon emulsion characteristics. International Journal of Food Science and Technology 37:263–269.

Downey, G., E. Sheehan, C. Delahunty, D. O’Callaghan, T. Guinée, and V. Howard. 2005.Prediction of maturity and sensory attributes of Cheddar cheese using near-infraredspectroscopy. International Dairy Journal 15: 701–709.

Duce, S.L., T.A. Carpenter, and L.D. Hall. 1992. Nuclear magnetic resonance imaging of freshand frozen courgettes. Journal of Food Engineering 16: 165–172.

Duce, S.L., M.H.G. Amin, M.A. Horsfield, M. Tyszka, and L.D. Hall. 1995. Nuclear magneticresonance imaging of dairy products in two and three dimensions. International DairyJournal 5: 311–319.


Dufour, É. and A. Riaublanc. 1997. Potentiality of spectroscopic methods for the character-isation of dairy products. I: Front-face fluorescence study of raw, heated and homogen-ised milks. Lait 77: 671–681.

Dufour, É., P. Robert, D. Renard, and G. Llamas. 1998. Investigation of b-lactoglobulingelation in water=ethanol solutions. International Dairy Journal 8: 87–93.

Dufour, É., M.F. Devaux, P. Fortier, and S. Herbert. 2001. Delineation of the structure of softcheeses at the molecular level by fluorescence spectroscopy—Relationship with texture.International Dairy Journal 11: 465–473.

Engelsen, S.B., M.K. Jensen, H.T. Pedersen, L. Nørgaard, and L. Munck. 2001. NMR-bakingand multivariate prediction of instrumental texture parameters in bread. Journal ofCereal Science 33: 59–69.

Freidman, H., J. Whitney, and A. Szcze�sniak. 1963. The texturometer: A new instrument forobjective texture measurement. Journal of Food Science 28: 390–396.

Fu, X., Y. Ying, H. Lu, H. Xu, and H. Yu. 2007. FT-NIR diffuse reflectance spectroscopy forkiwifruit firmness detection. Sensing and Instrumentation for Food Quality and Safety1: 29–35.

Gamble, G.R. 1994. Non-invasive determination of freezing effects in blueberry fruit tissue bymagnetic resonance imaging. Journal of Food Science 59: 573–610.

González, J.J., K.L. McCarthy, and M.J. McCarthy. 2000. Textural and structural changes inlasagna after cooking. Journal of Texture Studies 31: 93–108.

Hickey, H., B. MacMillan, B. Newling, M. Ramesh, P. van Eijck, and B. Balcom. 2006.Magnetic resonance relaxation measurements to determine oil and water content in friedfoods. Food Research International 39: 612–618.

Hills, B.P., S.F. Takács, and P.S. Belton. 1990. A new interpretation of proton NMRrelaxation time measurements of water in food. Food Chemistry 37: 95–111.

Ilker, R. and A.S. Szcze�sniak. 1990. Structural and chemical bases for texture of plantfoodstuffs. Journal of Texture Studies 21: 1–36.

Irie, K., A.K. Horigane, S. Naito, H. Motoi, and M. Yoshida. 2004. Moisture distribution andtexture of various types of cooked spaghetti. Cereal Chemistry 81: 350–355.

Jagannathan, N.R., R. Jayasundar, V. Govindaraju, and P. Raghunathan. 1994. Applicationsof high resolution magnetic resonance imaging (MRI) and spectroscopy (MRS)techniques to plant materials. Indian Academy of Sciences Chemical Science 106:1595–1604.

Jarvis, M.C. and D.C. Apperley. 1990. Direct observation of cell wall structure in living planttissue by solid-state 13C NMR spectroscopy. Plant Physiology 92: 61–65.

Karoui, R., G. Mazerolles, and É. Dufour. 2003. Spectroscopic techniques coupled withchemometric tools for structure and texture determinations in dairy products. Inter-national Dairy Journal 13: 607–620.

Kerr, W.L., C.J. Clark, M.J. McCarthy, and J. De Ropp. 1997. Freezing effects in fruit tissue ofkiwifruit observed by magnetic resonance imaging. Scientia Horticulturae 69: 169–179.

Lai, H.-M. and S.-C. Hwang. 2004. Water status of cooked white salted noodles evaluated byMRI. Food Research International 37: 957–966.

Lebecque, A., A. Laguet, M.F. Devaux, and É. Dufour. 2001. Delineation of the texture ofSalers cheese by sensory analysis and physical methods. Lait 81: 609–623.

Létal, J., D. Jirák, L. Šudelová, and M. Hájek. 2003. MRI ‘‘texture’’ analysis of MR images ofapples during ripening and storage. Lebensmittel-Wissenschaft und-Technologie 36:719–727.

Leung, H.K., J.A. Magnuson, and B.L. Bruinsma. 1979. Pulsed nuclear magnetic resonancestudy of water mobility in flour doughs. Journal of Food Science 44: 1408–1411.

Lu, R. 2007. Nondestructive measurement of firmness and soluble solids content for applefruit using hyperspectral scattering images. Sensing and Instrumentation for FoodQuality and Safety 1: 19–27.


Mazerolles, G., M.F. Devaux, G. Duboz, M.H. Duployer, M. Riou, and É. Dufour. 2001.Infrared and fluorescence spectroscopy for monitoring protein structure and interactionchanges cheese ripening. Lait 81: 609–623.

McCarthy, M.J., B. Zion, P. Chen, S. Ablett, A.H. Darke, and P.J. Lillford. 1995. Diamagneticsusceptibility change in apple tissue after bruising. Journal of the Science of FoodAgriculture 67: 13–20.

McCarthy, M.J., Y.J. Choi, A.G. Goloshevsky, J.S. De Ropp, S.D. Collins, and J.H. Walton.2006. Measurement of fluid viscosity using microfabricated radio frequency coils.Journal of Texture Studies 37: 607–619.

Mehinagic, E., G. Royer, D. Bertrand, R. Symoneaux, F. Laurens, and F. Jourjon.2003. Relationship between sensory analysis, penetrometry and visible-NIR spectros-copy of apples belonging to different cultivars. Food Quality and Preference 14:473–484.

Meullenet, J.-F., A. Mauromoustakos, T.B. Horner, and B.P. Marks. 2002. Prediction oftexture of cooked white rice by near-infrared reflectance analysis of whole-grain milledsamples. Cereal Chemistry 79: 52–57.

Miquel, M.E., S.D. Evans, and L.D. Hall. 1998. Three dimensional imaging of chocolateconfectionery by magnetic resonance methods. Lebensmittel-Wissenschaft und-Techno-logie 31: 339–343.

Mleko, S., E.C.Y. Li-Chan, and S. Pikus. 1997. Interactions of k-carrageenan with wheyproteins in gels formed at different pH. Food Research International 30: 427–433.

Ortíz, C., P. Barreiro, E. Correa, F. Riquelme, and M. Ruiz-Altisent. 2001. Non-destructiveidentification of woolly peaches using impact response and near-infrared spectroscopy.Journal of Agricultural Engineering Research 78: 281–289.

Povlsen, V.T., Å. Rinnan, F. van den Berg, H.J. Andersen, and A.K. Thybo. 2003. Directdecomposition of NMR relaxation profiles and prediction of sensory attributes of potatosamples. Lebensmittel-Wissenschaft und-Technologie 36: 423–432.

Primo-Martín, C., N.H. van Nieuwenhuijzen, R.J. Hamer, and T. van Vleit. 2007. Crystallinitychanges in wheat starch during the bread-making process: Starch crystallinity in thebread crust. Journal of Cereal Science 45: 219–226.

Qing, Z.S., B.P. Ji, and M. Zude. 2007a. Wavelengths selection for predicting physico-chemical apple fruit properties based on near infrared spectroscopy. Journal of FoodQuality 30: 511–526.

Qing, Z.S., B.P. Ji, and M. Zude. 2007b. Predicting soluble solids content and firmness inapple fruit by means of laser light backscattering image analysis. Journal of FoodEngineering 82: 58–67.

Quintana, J.M., A.N. Califano, N.E. Zaritzky, P. Partal, and J.M. Franco. 2002. Linear andnonlinear viscoelastic behavior of oil-in-water emulsions stabilized with polysacchar-ides. Journal of Texture Studies 33: 215–236.

Rondeau-Mouro, C., B. Bouchet, B. Pontoire, P. Robert, J. Mazoyer, and A. Buléon. 2003.Structural features and potential texturising properties of lemon and maize cellulosemicrofibrils. Carbohydrate Polymers 53: 241–252.

Ruan, R.R., C. Zou, C. Wadhawan, B. Martínez, P.L. Chen, and P. Addis. 1997. Studies ofhardness and water mobility of cooked wild rice using nuclear magnetic resonance.Journal of Food Processing and Preservation 21: 91–104.

Šáfá�r, M., P.R. Bertrand, M.F. Devaux, and C. Génot. 1994. Characterization of edible oils,butters and margarines by Fourier transform infrared spectroscopy with attenuated totalreflectance. Journal of the American Oil Chemists’ Society 71: 371–377.

Sônego, L., R. Ben-Arie, J. Raynal, and J.C. Pech. 1995. Biochemical and physical evaluationof textural characteristics of nectarines exhibiting woolly breakdown: NMR imaging,X-ray computed tomography and pectin composition. Postharvest Biology and Tech-nology 5: 187–198.


Sørensen, L.K. and R. Jepsen. 1998. Assessment of sensory properties of cheese by near-infrared spectroscopy. International Dairy Journal 8: 863–871.

Szcze�sniak, A. 1963. Classification of textural characteristics. Journal of Food Science 28:385–389.

Szcze�sniak, A. 1987. Correlating sensory with instrumental texture measurements: An over-view of recent developments. Journal of Texture Studies 18: 1–15.

Thebaudin, J.-Y., A.-C. Lefèbvre, and A. Davenel. 1998. Determination of the cooking rate ofstarch in industrial sauces: Comparison of nuclear magnetic resonance relaxometry andrheological methods. Sciences Des Aliments 18: 283–291.

Thybo, A.K., I.E. Bechmann, M. Martens, and S.B. Engelsen. 2000. Prediction of sensorytexture of cooked potatoes using uniaxial compression, near infrared spectroscopy andlow field 1H NMR spectroscopy. Lebensmittel-Wissenschaft und-Technologie 33:103–111.

Thybo, A.K., P.M. Szczypi�nski, A.H. Karlsson, S. Dønstrup, H.S. Stødkilde-Jørgensen, andH.J. Andersen. 2004. Prediction of sensory texture quality attributes of cooked potatoesby NMR-imaging (MRI) of raw potatoes in combination with different image analysismethods. Journal of Food Engineering 61: 91–100.

Thygesen, L.G., A.K. Thybo, and S.B. Engelsen. 2001. Prediction of sensory texture qualityof boiled potatoes from low-field 1H NMR of raw potatoes. The role of chemicalconstituents. Lebensmittel-Wissenschaft und-Technologie 34: 469–477.

Valéro, C., M. Ruiz-Altisent, R. Cubeddu, et al. 2004. Detection of internal quality in kiwiwith time-domain diffuse reflectance spectroscopy. Applied Engineering and Agricul-ture 20: 223–230.

van Dijk, C., M. Fischer, J. Holm, J.-G. Beekhuizen, T. Stolle-Smits, and C. Boeriu. 2002.Texture of cooked potatoes (Solanum tuberosum). 1. Relationships between dry mattercontent, sensory-perceived texture, and near-infrared spectroscopy. Journal of Agricul-ture and Food Chemistry 50: 5082–5088.

Villareal, C.P., N.M. De La Cruz, and B.O. Juliano. 1994. Rice amylose analysis by near-infrared transmittance spectroscopy. Cereal Chemistry 71: 292–296.

Wang, S.Y., P.C. Wang, and M. Faust. 1988. Non-destructive detection of watercore in applewith nuclear magnetic resonance imaging. Scientia Horticulturae 35: 227–234.

Wang, C.Y. and P.C. Wang. 1989. Nondestructive detection of core breakdown in ‘Bartlett’pears with nuclear magnetic resonance imaging. HortScience 24: 106–109.

Xie, F., F.E. Dowell, and X.S. Sun. 2003. Comparison of near-infrared reflectance spectros-copy and texture analyzer for measuring wheat bread changes in storage. CerealChemistry 80: 25–29.

Yakubu, P.I., E.M. Ozu, I.C. Bäianu, and P.H. Orr. 1993. Hydration of potato starch inaqueous suspension determined from nuclear magnetic studies by 17O, 2H, and 1HNMR: Relaxation mechanisms and quantitative analysis. Journal of Agricultural andFood Chemistry 41: 162–167.

Zude, M., I. Birlouez, J. Paschold, and D.N. Rutledge. 2007. Nondestructive spectral-opticalsensing of carrot quality during storage. Postharvest Biology and Technology 45: 30–37.

Zude, M., L. Spinelli, and A. Torricelli, 2008. Approach for nondestructive pigment analysisin model liquids and carrots by means of time-of-flight and multiwavelength remittancereadings. Analytica Chimica Acta 623: 204–212.


5.2 SPECTROSCOPIC TECHNIQUE FOR MEASURING THE TEXTUREOF HORTICULTURAL PRODUCTS: SPATIALLY RESOLVEDAPPROACH

RENFU LU

5.2.1 INTRODUCTION

Texture is an important component in evaluating the overall quality of fresh horti-cultural products. People with different cultural, ethnic, and age backgrounds wouldhave different expectations for the texture of individual horticultural products. Sincetexture is the human’s perception of the physical and mechanical properties of a foodby the feeling of touch, it cannot be accurately described using single well-definedengineering parameters. Despite the difficulty in defining texture, there are certaincommon textural properties that are considered important by consumers. Forexample, it is generally agreed that firmness is one of the most important texturalproperties of horticultural products. The importance of firmness is underlined by thefact that it is a key parameter in the grading standards for many horticulturalproducts. Like many terminologies used in describing food texture, firmness is anelusive term; it has different meanings to different people. Food sensory scientistswould define fruit firmness as the degree of force perceived by a person in crushing apiece of fruit using his or her molar teeth. Horticulturists, on the other hand, mayconsider the force required to penetrate a cylindrical probe into a fruit sample for aspecific depth to be a measure of firmness. Engineers may use elastic modulus orfailure strength to indicate the firmness of horticultural samples. Hence, it is notsurprising that many different methods have been used to measure the firmness ofhorticultural products. The Magness–Taylor (MT) firmness tester is the most widelyused in the horticultural industry and research laboratories.

MT firmness testing is performed using a cylindrical steel probe with a curvedend surface and a specific diameter to penetrate the flesh tissue for a specific depth,and the maximum force recorded during the penetration is used to indicate thefirmness of the product item. MT firmness measurements involve a complex formof mechanical load, including compression, shearing, and tension; they essentiallyreflect the composite mechanical failure strength of the test samples, which isdifficult to quantify using engineering mechanics theory. MT firmness measurementsare susceptible to operational error and may not yield consistent results (Harker et al.1996). Despite these drawbacks, the MT firmness tester and its variants are simpleand easy to use, and correlate well with the human perception of firmness (Bourne2002). Hence, the MT firmness tester has been long used as a standard method formeasuring, monitoring, and inspecting the quality of horticultural products duringharvest, postharvest handling and storage, and marketing. A new nondestructivetechnique under development is often evaluated against the destructive MT firmnessmeasurement. Researchers have attempted to develop and use more objective, easier-to-define firmness measurement methods to replace the existing MT testing method,but these alternative methods have not been widely adopted (Abbott et al. 1997).


Researche rs have explor ed and developed many nondestruct ive firmne ss meas-urement techni ques for horti cultural product s; most of them have not been used bythe industry eith er because they are not pract ical or because they cannot provi deaccurate, reliable measurem ent of firmne ss, especiall y when they are eva luatedagainst the destructi ve MT firmne ss meas uremen t (Ab bott et al. 1997). Earlyresearch was mainly focuse d o n n ondestruct ive mecha nical methods such as quasi-static force=deform ation, vibration , impact, a nd sonic resona nce (Ab bott et al. 1997 ;Lu and Abbo tt 2004). Among these , impact and sonic techni ques look p articularlypromisi ng for nonde structive evalua tion of the text ure of horticultural product sbecause they woul d cause littl e or no damage to product items and bec ause theyare rapid and suitable for online sorting and gradin g. Impa ct and sonic techniqueshave been used recent ly in some moder n commerci al packin ghouses for sort ing andgrading frui t. However , firmne ss measurem ents b y the imp act and sonic techniquestend to be in fluence d by frui t size and shape or geometry. The techniques seem towork better for certain type of frui ts, but they may not wor k well for firm productssuch as apple (Shmulevi ch and Howarth 2003). Moreo ver, since impact and sonictechniques essentially measure elastic param eters, they may no t correlate wi th MTfirmness well. Hence, resear ch on the develo pment of more effect ive nondest ructivefirmness measu rement techniques is continuin g to draw consi derabl e inte rest fromresearchers around the world.

With the rapid develo pment s in compu ter and optical techno logies, especi allythe detect ing d evices over the past decade, compu ter imaging and spectroscop y arenow being widely used for qu ality inspectio n of horti cultural and food products.NIR spect roscop y is one such prom inent techno logy that has recent ly gainedincreasing application s in moni toring, grading , and meas uring the quali ty of freshhorticultural product s. NIR spect roscop y measures an aggrega te amoun t of lightdiffusely re flected or trans mitted from a product item over a spectral range between780 and 2500 nm (mor e broadl y, it may also incl ude the v isible range betw een 400and 780 nm). According to the Beer –Lambert law (Williams and Norris 2001), theabsorption of light at a speci fi c wave length is proportion al to the concent ration of achemical compo nent in the samp le. Hence NIR spect ra are shaped by the absorp tionof chemical components in the product item at specific wavelengths or bands. Sincedifferent chemical components often absorb light at different wavelengths or bands,they could be measured simultaneously from NIR spectra.

Numerous studies have been reported of using NIR spectroscopy for assessingthe texture of food products. NIR spectroscopy, e.g., has been used for measuringthe hardness of wheat kernels, the texture of bread, and the tenderness of meat(see Sectio n 5.1 for an overvi ew of NIR and other optic al techno logies formeasuring the texture of foods). Studies have also been reported of using NIRspectroscopy to measure the firmness of fresh horticultural products (Lammertynet al. 1998; McGlone and Kawano 1998; Lu et al. 2000; Lu 2001). These studieshave showed that NIR technology is still not viable for evaluation of the firmnessof fresh horticultural products. Such finding is not surprising in view of thecomplexity of firmness measurement and the underlining principle of NIRmeasurement. Firmness reflects the composite mechanical failure strength for a


product item, and is determined by its structural and physical properties, which arein turn related to the physiological activities and maturity level. When light entersa turbid material, scattering and absorption occur. Absorption is primarily relatedto the chemical properties of the material, whereas scattering is influenced by thedensity and structural characteristics. Scattering is dominant in turbid biologicalmaterials over the visible and short-wave NIR region of 500–1300 nm. Hencemeasurement and separation of the scattering and absorption properties couldprovide more useful information on the chemical composition and structuralcharacteristics of a horticultural product. Since conventional NIR spectroscopycannot measure and separate absorption and scattering, its ability for assessingfirmness is restricted.

Considerable research has been reported on the measurement of the opticalproperties of human tissues as a diagnostic tool for medical applications (Tuchin2000; Vo-Dinh 2003). Several measurement techniques have been developed, whichinclude spatially resolved, time-resolved, and frequency-domain techniques (Tuchin2000). But only limited research has been reported on the measurement of the opticalproperties (absorption and scattering) of food and agricultural products (Birth 1978;Birth et al. 1978; Cubeddu et al. 2001; Qin and Lu 2006; Xia et al. 2007). This isbecause the techniques for measuring the optical properties are still not well devel-oped, and are too expensive and not sufficiently reliable and robust for food andagricultural products. Hence, much research is needed in the development of apractical, cost-effective technique for measuring the optical properties of food andagricultural products, quantification of the light–plant tissue interactions, and estab-lishment of the relationship between the measured optical properties and qualityattributes or properties of these products.

This section reviews the latest developments and applications of spatiallyresolved spectroscopic techniques for measuring the optical properties and textureor firmness of horticultural products. The section first provides a brief overview ofthe theory of light transfer in turbid biological materials and the principle of spatiallyresolved techniques for measuring the optical properties. It then presents a hyper-spectral imaging technique along with the data analysis methods and procedures fordetermining the optical properties of horticultural products and for quantifying fruitfirmness and light propagation in apples. Thereafter, the section reviews the latestdevelopments of a light-scattering technique and mathematical methods for charac-terizing the spectral scattering profiles of horticultural products to assess fruitfirmness. The section ends with some concluding remarks on future research needsfor these emerging technologies.

5.2.2 LIGHT PROPAGATION IN SCATTERING-DOMINANT BIOLOGICAL MATERIALS

5.2.2.1 Scattering and Absorption

Most biological materials including food and agricultural products are opticallyopaque in the visible and NIR region between 500 and 1300 nm because photonstraveling within them will undergo multiple scattering. Scattering is a physical


phenomenon that causes change in the traveling directions of photons in the turbidmaterial without loss of their electromagnetic energy, i.e., photons remain the samewavelength or the same level of energy after scattering. Biological materials areheterogeneous at the microstructural level, with individual components being differ-ent in their optical properties. Scattering will occur in the form of refraction or theredirection of a light wave that is propagating from one material or component to asecond material with a different refractive index. In a scattering-dominant material,photons will undergo multiple scattering before being absorbed or reemerging fromthe material. Scattering depends on the density and structural characteristics of thebiological material, and hence it could be used for ascertaining its physical charac-teristics and properties.

Absorption, on the other hand, is a process involving the extraction of energyfrom photons by the molecules and atoms of the turbid material, which is subse-quently converted into other forms of energy (such as heat and photochemicalenergy) or is released in the form of lower energy (longer-wavelength) light suchas fluorescence. According to the quantum theory, atoms and molecules only absorbphotons in specific transition, and the absorbed energy is used to increase theirinternal energy states. The regions of the spectrum where this energy absorptionoccurs are known as absorption bands; these bands are specific to particular molecu-lar species. Absorption of photons by molecules and atoms may take place in threedifferent patterns: (1) electronic, (2) vibrational, and (3) rotational. Electronic tran-sitions occur in both molecules and atoms, whereas vibrational and rotationaltransitions occur in molecules. NIR spectroscopy is primarily based on the funda-mental vibrational transition mechanism.

Consider a simple case, in which a collimating beam is incident upon a layer ofhomogeneous medium of thickness (L) [cm], the differential change of light intensity(dI) of the collimating light beam traversing an infinitesimal distance (dz) through themedium may be described as follows:

dI ¼ �(ma þ ms)I dz ¼ �mtI dz (5:2:1)

whereI [W cm�2] is the intensity of transmitted lightma [cm

�1] is the absorption coefficientms [cm

�1] is the scattering coefficientmt [cm

�1] is the extinction coefficient (or total attenuation coefficient)

Integrating over the layer’s thickness L gives the well-known Beer–Lambert law:

I ¼ I0 exp (�mtL) (5:2:2)

which may also be expressed as

I ¼ I0 exp (�«ciL) (5:2:3)


whereI0 [W cm� 2] is the inci dent inte nsity« is mol ar absorptivityci is mol ar concent ration of the absorb er

Transmiss ion, T , is a commonl y used quanti ty, whi ch is defined as the ratio oftransmitte d intensity I to incid ent densi ty I0,

T ¼ I

I0( 5: 2:4)

The absorb ance ( A ) or optical density (OD) of an attenuati ng medi um is given by

A ¼ OD ¼ log10I0I

� �¼ log10

1T

� �¼ «ci L ( 5: 2:5)

Equation 5.2.5 is the basis of moder n NIR spectroscop y, whi ch states that theconcentrati on of an absorb ing constitu ent is line arly propor tion al to the absorb anceor the logarithm of transmi ssion. Stric tly speaking, Equ ation 5.2.5 is only validfor one-di mensional cases in which the inci dent light beam is coll imated. MostNIR applicati ons are implem ented in one of the three sensi ng modes : diff userefl ectanc e, interacta nce, and transmi ssion (Ab bott et al. 1997; Sc haare andFraser 2000). When re fl ectance (or transmi ssion or interacta nce) meas urementsare perfor med on an intact product item in which scatt ering is dominant , Equ ation5.2.5 can only provide an estimate of total light attenuation in the productitem, and it cannot be used to determine ma and m s separa tely (see also Sections1.3.1 an d 1.3.2).

5.2.2.2 Diffusion Theory Model

Light interaction with a turbid medium is rather complicated. A rigorous treatment ofthe light propagation in a scattering medium will require invoking the radiationtransfer equation (also known as the Boltzmann equation), which can be derivedfrom the balance of energy flowing into and out the scattering medium (Tuchin2000). The radiation transfer equation, which is expressed in a general integro-differential form, is difficult to solve analytically.

For many biological materials, scattering is dominant over the visible andshort-wave NIR region (approximately 500–1300 nm). Under the assumption ofscattering dominance (ms�ma), the transport of photons in the turbid material maybe approximated to be a diffusion process. The radiation transfer equation can thenbe expressed by an approximate diffusion theory model, which is given as

1c

dF(~r,t)dt

¼ Dr2F(~r,t)� maF(~r,t)þ S(~r,t) (5:2:6)


wherer2 is the Laplaci an op eratorc [m s� 1] is the veloci ty of light in the mediumF(~r ,t ) ¼ Ð

4 p I (~r , ~V,t ) dV is the photon densi ty (also called the flu ence rate

[phot ons cm� 2 . s�1 ])D ¼ [3( ma þ m0

s )] � 1 is the diff usion coef ficient

S(~r , t ) is the source in the scatt ering medium that gives the number of photonsemi tted at position ~r and time t per unit volume per unit time. m 0s is calledthe reduced scatteri ng coef ficient, whic h is related to the scatt ering coef fi-cient (als o called the aniso tropic scatteri ng coef fic ient), by the follow ingrelat ionship:

m0s ¼ ( 1 � g )m s ( 5: 2:7)

where g is the aniso tropy facto r that indi cates the numbe r o f scatt ering eventsrequired before the initial photon propaga tion direc tion is comple tely randomized .

The value of g ranges from �1 to 1; g ¼ 0 represents isot ropic scattering, g ¼ 1represents total forward scatte ring, while g ¼�1 repres ents tota l backward scatter-ing. For most biological materials, the value of g is between 0.60 a nd 0.99.

The rate of incre ase of photon s (the left-hand side of Equ ation 5.2.6) wi thin asample volume is equal to the number of photon s scattered into the volume elemen tfrom its surro unding s (the first term of the right -hand side of Equati on 5.2.6) minusthe numbe r of photon s absorb ed within the volume e lement (the second term of theright-hand side of Equati on 5.2.6 ), plus the numbe r of photon s emitted from sourceswithin the volume elem ent at posit ion ~r and t . With the know ledge of ma an d m 0scoupled with appropr iate boundar y cond itions, we may solve Equa tion 5.2.6 eitheranalytical ly or numerical ly to quantify ligh t propaga tion o r photon distrib utions in aturbid biological mat erial.

Analyti cal solution s of the approxi mat e diffusion theor y model have beenobtained for simple geome tries of turbid media, such as slabs and semi-in fi nitemedia, for d ifferent forms of light source s (i.e., stead y-state or conti nuous-wav e[cw] light, pulsed light, and intensity-m odulat ed ligh t). The next subsec tion presen tsa solution for a steady-state light illumination case that is of great practical interestand is directly related to the spatially resolved hyperspectral imaging techniquedescribed in Section 5.2 .3 for meas uring the opti cal proper ties of food and agric ul-tural products.

5.2.2.3 Steady-State Solutions

Conside r a semi-in finite homog eneous turbid medium as show n in Figure 5.2.1, inwhich scattering is dominant (m0

s �ma). A steady-state (or cw) point light isimpinged vertically on the surface of the medium. Assume that no photon sourceexists in the medium, S(~r, t)¼ 0. Then the approximate diffusion theory model(Equation 5.2.6) is reduced to


Dr2F(~r,t) ¼ maF(~r,t) (5:2:8)

A solution to Equation 5.2.8 for the problem described has been derived by Farrellet al. (1992). The diffuse reflectance at the surface of the semi-infinite medium,which can be derived from the photon fluence rate (F), is expressed as a function ofdistance from the light source:

Rf r;ma,m0s

� � ¼ a0

4p1m0t

meff þ1r1

� �exp(�meffr1)

r21

�

þ 1m0t

þ 4A3m0

t

� �meff þ

1r2

� �exp(�meffr2)

r22

�(5:2:9)

wherer is the distance from the incident pointa0 is the transport albedo, a0 ¼ m0

s=(ma þ m0s)

meff is the effective attenuation coefficient, meff ¼ [3ma(ma þ m0s)]

1=2

m0t is the total attenuation coefficient, m0

t ¼ ma þ m0s

The variables r1 and r2 are given by the following two equations:

r1 ¼ 1m0t

� �2

þ r2" #1

2

(5:2:10)

Reflectance, Re(r )Continuous-wave beam

r

n1,

FIGURE 5.2.1 Principle of the steady-state spatially resolved technique for measuring theoptical properties of biological materials.


r2 ¼ 1m0t

þ 4F3m0

t

� �2

þ r2" #1

2

(5:2:11)

where F is an internal reflection coefficient determined by the mismatch of relativerefractive index (nr) at the interface, and it may be calculated by the followingempirical equation (Groenhuis et al. 1983):

F ¼ 1þ rd1� rd

(5:2:12)

in which

rd � �1:44n�2r þ 0:710n�1

r þ 0:668þ 0:0636nr (5:2:13)

nr ¼ nsnair

(5:2:14)

The specific refractive index for turbid biological materials (ns) varies slightly withwavelength; however, in many practical applications, we may treat ns constant(Nichols et al. 1997; Dam et al. 1998). Thus, once ma and m0

s are known, the shapeof the spatial reflectance profile is uniquely determined by Farrell’s diffusion theorymodel (Equation 5.2.9). Conversely, if the reflectance profile resulting from a pointlight source over the surface of the turbid medium is measured, ma and m0

s may bedetermined by applying an inverse algorithm to Equation 5.2.9. Farrell modelprovides an excellent description of the spatial diffuse reflectance profiles for turbidbiological materials in which scattering is dominant (Farrell et al. 1992; Nichols et al.1997; Dam et al. 1998; Gobin et al. 1999). It provides the mathematical basis for thespatial resolved technique that is described in Section 5.2.3.

5.2.3 HYPERSPECTRAL IMAGING TECHNIQUE FOR MEASURING THE OPTICAL

PROPERTIES OF HORTICULTURAL PRODUCTS

Hyperspectral imaging is a relatively new technology that has emerged as a powerfulmethod for quality evaluation and safety inspection of food and agricultural productsin the past decade. It possesses the features of imaging and spectroscopy, and thusenables us to obtain both spectral and spatial information from an object. As a result,the technique is particularly useful for detecting quality attributes and chemicalcomponents in a product item that are spatially variable and hence may be difficultto ascertain using either imaging or spectroscopic techniques. Hyperspectral imaginghas been used for detecting quality of fruit (such as bruises and sugar distributions inthe fruit) and diseases or wholesomeness of poultry and meat products (Lu and Chen1998; Martinsen and Schaare 1998; Lu 2003; Park et al. 2002). In this section, wepresent a novel application of hyperspectral imaging technique for measuring theoptical properties of horticultural products and for assessing fruit texture or firmness.


5.2.3.1 Principle and Instr umentat ion

The meas urem ent of the opti cal proper ties of horticult ural products by the hyper-spectral ima ging techni que is based on the spati ally resolved (SR) spect roscopicprinciple , which is schema tically shown in Figure 5.2.1. As a small, coll imatedbroadba nd beam is incident up on the surfa ce of a semi -infin ite turbid medium, thelight is scattered into different direction s, and some ligh t is absorb ed. A fract ion ofthe light will backsca tter and exit from the surface, generat ing diffuse re flectanc e atthe surfac e of the medium. By capturing the re flectanc e pro files from the samplesurface using an optical device for indi vidual wave lengths, we may deter mine theoptical properties (ma an d m 0s ) using an inver se algorithm for Farrel l’ s diffusiontheory model (Equation 5.2.9) to fit the indi vidual scatt ering pro files. Clearly, tomeasure the optical proper ties over a range of wave lengths using a spatiall y resolvedmethod, one needs an opti cal system that is cap able of acquir ing both spect ral andspatial informat ion from the sample. Hype rspectral ima ging is thus ideally suited toaccomplis h this task.

The hyperspec tral ima ging technique is commonl y imp lemented in one of thetwo sensing modes : (1) the line-scann ing or push-b room mode and (2) the band-pa ssfilter-base d mode. A line -scann ing hypers pectral ima ging system scans the objectone line at a time. Each pixel from the scanning line is represen ted as a spectrum onthe a rea-array charge- couple d device (CCD) detect or. A 3-D hypers pectr al imagecube is create d by sequent ially scanni ng the enti re surfa ce of the samp le. A filtered-based hypers pectral imaging syst em is comm only equipped with a liquid crystaltunable filter (LCTF) or an acousto-opt ic tunable filter (AOTF) to acquir e 2-D spatialimages for indi vidual wave lengths or bands in sequenc e. Eac h imaging mode hasits meri ts and shortcoming s. In the line-scann ing mode, there shoul d be relativemovement between the sample and the imaging syst em to obtain 3-D hyper-spectral images for the entire ob ject. This mode is advant ageous for o nlineimplemen tation because the movi ng object will presen t itself to the hypers pectralimaging syst em for the sequential scanni ng of its entire surfa ce. The filter-base dhyperspec tral imaging system, on the other hand, does not requi re the relativemovement between the samp le and the ima ging system. However , since 3-Dhyperspec tral ima ges are created via sequent ial acquis itions of spect ral imagesfor each wave length, this sensing mode is not ap propriate for real-tim e, onli neapplicatio ns. Moreover, spect ral image cali brations for the filter-base d mode canbe more complicated than those for the line-scanning mode. This section describes aline scanning-based hyperspectral imaging system for acquiring spatially resolvedspectral scattering profiles from intact fruit and vegetable samples, and appropriatemethods and algorithms for extracting the optical parameters from the scatteringprofiles.

A schematic of a hyperspectral imaging system for measuring the opticalproperties of horti cultural products is shown in Figure 5 .2.2. The system consi stsof three main units: a hyperspectral imaging unit, a light source unit, and a samplehandling unit. The hyperspectral imaging unit has a high-performance CCD camerawith a control device, an imaging spectrograph that line scans the sample anddisperses the light from the scanning line into different wavelengths while preserving


Camera controlCCD

camera

Imagingspectrograph

Zoomlens

Computer

Top viewScanning

lineIncident

light

Scanning area(10 scans, 9 mm)

Sample moving direction

Microlens

Optical fiber

Photoelectricsensor

Lamphousing Horizontal

stageFruit

Throughbeam

Motor1

Motor2

Vertical stageLamp control

Lamp

FIGURE 5.2.2 Schematic of a hyperspectral imaging system for acquiring spatially resolvedbackscattering images from a horticultural sample. (From Qin, J., Ph.D. dissertation, MichiganState University, East Lansing, MI, 2007.)

its original spatial information, and a zoom lens. To acquire high-quality hyperspec-tral images, it is important that the CCD camera have a large dynamic range (a 16-bitformat is preferred) with good SNRs and high photon efficiencies over thewavelength range of interest. The hyperspectral imaging unit needs to be calibratedboth spectrally and spatially. If spectral and spatial distortions (i.e., spectraland spatial lines on the image are not straight and=or parallel) occur to the hyper-spectral imaging unit, it will require full-scale spectral and spatial calibrations pixelby pixel. The procedure of performing full-scale spatial and spectral calibrations canbe quite complicated and time consuming (Lawrence et al. 2003). If spectraland spatial distortions are small (i.e., no more than one pixel) or negligible, asimpler procedure for spectral and spatial calibrations should be sufficient (Lu andChen 1998).

The light source unit consists of a lamp housing installed with a broadband lampoperating in cw mode, a computer-controlled device to provide a highly stable directcurrent (DC) output to the lamp, and an optic fiber assembly composed of an opticfiber and a collimating lens. A quartz tungsten halogen lamp is presently a preferredchoice because it has smooth spectral responses over the visible and NIR region.A DC-regulated light is required to maintain a stable light output. In addition, a lightintensity feedback control device is recommended so that the output level of the light


can be moni tored and adjusted in real time for high reprod ucibili ty (Peng and Lu2006a). The optic fiber assem bly delivers a ligh t beam to the samp le. The light beamshould be smal l (not more than 1.5 mm in diameter ) so that the assumpti on made inderiving Fa rrell ’s diffusion theor y model (Equ ation 5.2.9 ) is not violated. Finally, thesample-ha ndling unit moves the samp le to the desired position for the ima ging unitto take images at a speed that is synchr onized wi th the ima ging acquis ition speed.

After the ligh t beam enters the sample, a diffuse re flectanc e image will begenerated at the surfa ce of the samp le as a resul t of light scatteri ng and propaga tion.If the samp le is homogen eous and suf ficiently large in size, the diff use re flectanc eimage will be symmet ric to the be am incident point. In this case, we may takeadvantage of the symm etry featu re by only taking single line scans, inst ead ofscanning the entire scatteri ng surfa ce of the samp le. This arrang ement will onlycreate one particula r 2-D scatteri ng ima ge for each samp le, and thus can saveconsiderabl e time in the ima ge acquisition.

To accurately meas ure the opti cal param eters, it is necess ary to perfor m instru-ment respon se cali brations to correct any nonuni form respon ses that may occur in thehyperspec tral ima ging system. Nonu niform inst rument respon ses refer to the phe-nomenon in which the light inte nsities record ed on each row of CCD p ixels(correspondi ng to a speci fic wave lengt h) are not equal when a scanni ng image istaken from a unifo rm flat surface that is evenly illuminate d. The y are caused by theimperfect ion or speci fic sett ings of the optic al compo nents (e.g., uneven respon ses ofthe CCD pixel s, a minor varia tion of the slit of the ima ging spect rograp h, and thevarying v iewing angle of the lens) o r the light source fluctuat ion. In most hyper-spectral ima ging appli cations, we can correct this nonuni form respon se probl em bycollecting a scanning image from a reference panel wi th the unifo rm spectralresponses over the spect ral regio n of inte rest under the same lightin g condition ,and then calculati ng the relative re flec tance between the samp le and the refere nce onthe pixel-by-p ixel basis. However , this correct ion approac h is inappropri ate for thehyperspec tral imaging system show n in Figure 5.2.2, because o f its speci al ligh tingarrangement and the more stringent requireme nts for nonuni form response correc-tions. Nonu niform respon ses are inst rument speci fic and may change as the sett ingsfor the ima ging syste m change. Figure 5.2.3 shows nonuni form response curves forthree wavelengths obtained using the hyperspectral imaging system shown in Figure5.2.2. The imaging system exhibited a considerable degree of nonuniform response;the nonuniformity increases to as high as about 40% at a distance of 20 mm from thebeam incident point. The nonuniform instrument response corrections should beperformed, using a standardized laboratory setup, for all spatial distances (rows ofpixels) that are used in the acquisition of scattering images for all wavelengthscovered by the imaging system (Qin 2007).

After the nonuniform instrument response corrections have been completed, theimaging system should be tested against the reference samples with known opticalproperties to calibrate the system on reproducible values. Depending on applicationneeds, either liquid or solid reference samples (phantoms) may be used. Liquidphantoms are commonly made of absorbing solution (such as black ink or otherdyes) and scattering material (such as Intralipid, an emulsion made of fat that would


00.6

0.7

0.8

0.9

1.0

550 nm650 nm750 nm

5 10Distance [mm]

Nor

mal

ized

inte

nsity

[a.u

.]

15 20

FIGURE 5.2.3 Normalized nonuniform instrument response as a function of distance fromthe light beam center for the hyperspectral imaging system at three selected wavelengths.(From Qin, J. and Lu, R., Appl. Spectrosc., 61, 388, 2007a.)

act as pure scatteri ng material s). Su ch liquid reference samp les are easy to produce ,but they may not resemble the g eometry of actual samples to be meas ured. So lidphantoms can be prepared to resemble actual samples of complex geometry=structure,and they can b e made of eith er trans parent medium (such as polymers, silicone, orgelatin) or inher ently scatteri ng material s like wax. Ideal ly, phanto ms should beprepare d so that they would resemble actual samples to be meas ured in shape, size,and other physical charact eristics . At least a few of these refere nce samp les shoul d beprepare d to cover a typi cal range of optical proper ties that would be expect ed fromthe actual samp les to be meas ured (Tuchin 2000).

Once the hypers pectr al imaging system is validated wi th phanto ms, it can thenbe used for measu ring the optical properties of real- world samples. To have betterSNRs and to minimize the effect of local tissue proper ty varia tion on the scatte ringimage acquisition, multip le scatteri ng ima ges shoul d be co nsidered. The hype rspec-tral ima ging unit shown in Figure 5.2.2 would acquir e up to 10 scatt ering imagesfrom each frui t sample and these images are then averaged to obtai n one averagescatteri ng ima ge.

5.2.3.2 Procedures of Determi ning the Op tical Propertie s

Figure 5.2.4 show s a hypers pectral scatteri ng ima ge acquired from a ‘ GoldenDelicious ’ apple, which is displ ayed in both 2-D (Figure 5.2.4a) and 3-D(Figure 5.2.4b) formats. The useful spect ral range was between 500 and 1000 nmand the spatial distance for each side of the scatteri ng pro file was 10 mm. A linetaken from the ima ge in the horizontal direction represents a spatial scattering pro filefor a specific wavelength (Figure 5.2.4c), whereas a vertical line taken from theimage represents a spectrum for a pixel from the scanning line at a specific distance


Distance [mm]

Distance [mm]

Distance [mm]0 10 20 25

Inte

nsity

[a.u

.]

1(c) (d)

(a) (b)

10�2

10�1

100

2 3 4 5 6 7

600 nm

800 nm675 nm

8 9 10 500

�25�20 �10

5000

0

0.5

0.5

1.0

1.0

1.5

1.5

2.0

2.0

2.5

2.5

�104

�104

600

600

400

200

700

700

�10�5

05

10

800

800

Wavelength [nm]

Wavelength [nm]

Wav

elen

gth

[nm

]

Inte

nsity

[CC

Dco

unt]

Inte

nsity

[CC

Dco

unt]

900

900

1000

1000

1200

1100

1100

FIGURE 5.2.4 Original hyperspectral scattering image for an apple fruit in (a) 2-D format(b) 3-D format, (c) the spatial scattering profiles at three wavelengths, and (d) a spectral profiletaken at the distance of 1.6 mm to the beam incident point. (From Qin, J., Ph.D. dissertation,Michigan State University, East Lansing, MI, 2007.)

from the light incident center (Figure 5.2.4d). Hence the hyperspectral image inFigure 5.2.4a may be viewed as either composed of hundreds of spectra, eachrepresenting a pixel of a different distance from the scanning line, or composed ofhundreds of spatial scattering profiles for different wavelengths. The optical param-eters will be determined from each spatial scattering profile (Figure 5.2.4c).

The following stepwise protocol was used to determine the optical properties offruit samples (Figure 5.2.5):

Originalhyperspectral

images

Correctedhyperspectral

images

Enhancedhyperspectral

images

Originalspatial profiles

Correctedspatial profiles

Instrumentresponsecorrection

Imagepreprocessing

Profilesextraction

Fruit sizecorrection

Normalizationand

curve fittingma and m�s values(500–1000 nm)

FIGURE 5.2.5 Procedure for determining the optical properties from the hyperspectralscattering images of horticultural samples.


1. Correct ing the original hyperspe ctral ima ge for the nonuni form instrum entresponses for eac h wave length.

2. Preproce ssing the correct ed image (e.g., the averagi ng of severa l row s ofpixels) to enhance image quali ty.

3. Extracting spati al scatteri ng pro files from the enhanced ima ge and reducingthe tw o-sided spati al scatt ering pro files to one-sided scattering pro files (byutilizing the symm etry featu re).

4. Correcting the scattering profiles, as needed, for the effect of fruit size (seethe detai led descri ption of this step in Secti on 5 .2.4.1).

5. Performing normalizations for the scattering profiles (see further discussionbelow) and determining the values of ma and m0

s for each wavelength via anonlinear curve-fitting algorithm to fit each normalized scattering profilewith Equation 5.2.9.

The spatial scattering profiles extracted in step 3 do not represent the actualabsolute reflectance intensities from the sample, and they are not on the same scale asthose described by Farrell’s diffusion theory model (Equation 5.2.9). Hence it isnecessary to normalize both experimental scattering profiles (designated as Re) andFarrell model (designated as Rf) with respect to a specific scattering distance. Thenormalized experimental and Farrell’s scattering profiles are given by the followingequations:

Re,n(r) ¼ Re(r)

Re(rnormal)(5:2:15)

Rf,n(r) ¼ Rf (r)

Rf (rnormal)(5:2:16)

whereRe,n and Rf,n are the normalized experimental and Farrell model profiles,respectively

rnormal is the distance chosen for normalization (it is convenient to use the pointclosest to the light incident center)

The normalization process avoids the need of measuring absolute reflectance pro-files, which are more difficult to measure experimentally.

In performing the nonlinear curve-fitting procedure in step 5, a three-stepprocedure is recommended. The procedure includes (1) treating both ma and m0

s asunknown variables and obtaining their values through the nonlinear curve-fittingalgorithm; (2) fitting the m0

s values obtained in the first step with the followingwavelength-dependent function:

m0s ¼ al�b (5:2:17)

where a and b are parameters for the power series model, and they are related to thedensity and the average size of the scattering particles, respectively (Mourant et al.


1997); and (3) inser ting the results of m0s from step 2 into the normaliz ed Farrell

model, an d repeat ing the same fitting procedu re in step 1 with ma as the onlyunknown. This three -step curve- fitting procedu re would reduce the fitting no ise forboth ma and m 0s , thus imp roving the accuraci es over the one-step procedu re (Qin andLu 2007a).

5.2.4 A PPLICATIONS

In this secti on, severa l applicati on examp les of the spatiall y resolv ed hypers pectralimaging techni que are given, which include the measurem ent of the optical proper -ties of selec ted fruits and vegeta bles, e valuation of the fi rmness of app le fruit, andquantitative analys is of light propaga tion regard ing the penetr ation depths and lightdistributio ns in apple frui t.

5.2.4.1 Optical Properti es of Fruit s an d Veg etable s

Measureme nt of the opti cal proper ties of horticu ltural product s can help us betterunderstand the light –tissue inte ractions and develo p more effect ive optical sensingtechniques for asses sing product quali ty. Hor ticultural product s vary in size andshape, and their surfa ce usual ly is not flat. Henc e, there will be limit ations inapplying the diff usion theory model to horti cultural product s, and certa in assump-tions and sim plificati ons are needed. In a pplying Fa rrell ’s diff usion theory model(Equation 5.2 .9) to horti cultural products, we assum e that the samp les are homo-geneous in stru cture and suf ficient ly large in size compa red with the ligh t-scatte ringdistances so that they may be consi dered as a semi-in finite medium. However , actualhorticultural products are inhom ogeneous , and they have a prote ctive surfa ce layer(skin) whose stru ctural charact eristics or proper ties are quite diff erent from those ofthe interior tis sue. Henc e, in using the Farrell model to deter mine ma an d m 0s , weimplicitly assume that the opti cal proper ties meas ured are the combi nation of thoseof the skin and the flesh. It rema ins to be furt her inves tigated to what extent eachcompone nt co ntributes to the measured op tical proper ties or if a more compl exdiffusion model for multila yered medi a shoul d be employed. Despit e these limita-tions, the homogeneity assumption should provide a reasonable start in measuringthe optical properties of horticultural products.

Although it seems reasonable to treat fruits and vegetables (apple, peach, pear,etc.) as a semi-infinite medium for the purpose of using Farrell’s diffusion model, theactual effect of the curved sample surface on the measurement of diffuse reflectancemay not be ne glected. Figure 5.2. 6 show s how a curved surfa ce would causedistortions on the measurement of diffuse reflectance profiles from a sample ofcircular shape. Because of the finite distance between the imaging device and thesample, the actual acceptance angle for each point at the curved surface varies withthe distance from the central axis. As a result, the measured diffuse reflectance wouldunderestimate the actual reflectance, which becomes more significant as the distancer increases or if the size of samples becomes smaller. Hence, the intensities of themeasured reflectance should be corrected to take into account the effect of individual


Zoom lens

rz

L

I

2q0

hr

g

ra

Fruit

q

a

q1q2

a

b

Iv

FIGURE 5.2.6 Effect of fruit size and scattering distance on the measurement of diffusereflectance profiles (I, the reflectance component at an angle of u from the normal direction; Iv,reflectance component in the normal direction; L, distance between the sample and imaginglens; rz, lens radius; r, horizontal distance from the original axis; ra, fruit radius; u0, one-halfacceptance angle at the original point r¼ 0; u1¼a� g; u2¼aþb; a¼ asin(r=ra); b¼ atan[(rzþ r)=(Lþ h)]; g¼ atan[(rz� r)=(Lþ h)]; h¼ ra� [ra

2� r2]1=2).

samples’ size. The shape of many fruits and vegetables may be considered circular orspherical. Under this simplification, we can calculate the actual diffuse reflectancefrom the measured diffuse reflectance using the geometric relationship shown inFigure 5.2.6. Kienle and coworkers (1996) showed that the angular diffuse reflect-ance intensity for a scattering-dominant material obeys the Lambertian cosine law(Kortüm 1969). This implies that the reflectance component I with an angle u withrespect to the surface normal can be calculated as I¼ Ivcosu. For a spatial point witha distance of r from the light incident center, the reflectance measured by the imagingsystem is equal to the integration of the reflectance I over the acceptance angle(u2� u1) of the zoom lens (Figure 5.2.6), and it can be calculated by the followingequation (Lu and Peng 2007):

Re(r) ¼ðu2u1

IvdS cos2 u du ¼ Iv dS

u22þ sin 2u2

4

� �� u1

2þ sin 2u2

4

� �� (5:2:18)


wheredS repres ents a smal l line element that is perpend icul ar to the norm al direc tionu1 ¼ a � gu2 ¼ a þ ba ¼ asin( r=ra)b ¼ atan[( rz þ r)=( Ls þ h)]g ¼ atan[( rz � r)=( Ls þ h)]h ¼ ra � ( r a

2� r 2)1=2 (see Figure 5.2.6 for the meani ng of individua l symbols)

Equati on 5.2.18 shows that once the setting of the imaging syst em (i.e., thedistance between the samp le and the lens and the lens size) is chosen, the measuredrefl ectanc e Re would be dependen t on the scatteri ng dist ance and the curvat ure of thefruit. The acce ptance an gle for the meas ured re flectanc e at scatteri ng distance r ¼ 0 isequal to 2u0, where u0 ¼ a tan( r z=L) . If the scatt ering distance is greater than zero, theacceptanc e angle will change, whic h will in turn a ffect the meas ured re flectanc e.Ideally, the imaging syst em should collect the re flectanc e coveri ng the same ac cept-ance angle of 2 u0 at each scatteri ng distance. Thi s cannot be achiev ed with thecurrent imaging system. However , we may calculate the correct or actual re flectanc eat any scatteri ng distance based on the know ledge of the normal re flectanc e co m-ponent Iv, which is given by the follow ing equati on:

R (r ) ¼ðu 0

� u0

Iv d S cos 2 u d u ¼ I v d S u0 þ sin 2u0

2

� �¼ E ( r ) Re ( r ) (5: 2:19 )

where E( r ) is the correct ion facto r given by

E(r) ¼u0 þ sin 2u0

2u22þ sin 2u2

4

� �� u1

2þ sin 2u1

4

� � (5:2:20)

Equation 5.2.19 was used to correct the measured reflectance profiles for all testsamples before the procedure of determining the ma and m0

s values was started.Figure 5.2.7 shows the spectra of ma and m0

s from one sample for each of the ninefruits and vegetables over the spectral region of 500–1000 nm. It should be notedthat the spectra of ma and m0

s presented in Figure 5.2.7 do not necessarily representtypical value ranges for these products. Great variations in the optical properties existamong samples of the same type of products or even the same variety as shown laterfor ‘Golden Delicious’ apples. Nevertheless, these spectra reflect the general trendsand major features for these horticultural products. The absorption spectra forthe three varieties of apple (‘Golden Delicious’, ‘Fuji’, and ‘Red Delicious’), thepeach, the pear, and the kiwifruit all peaked at 675 and 970 nm, which wereattributed to the absorption by chlorophyll and water in the fruit tissue, respectively.The kiwifruit had the highest chlorophyll absorption, which could have beenexpected from its greenish flesh. The white flesh peach had the lowest value of theabsorption coefficient among the five fruit samples (Figure 5.2.7a), which could be


5000 0

5

10

15

0.2

0.4

0.6

0.8

1.0

1.2

600 700 800

Apple (RD)

Zucchini squashCucumberPlum

Wavelength [nm]900 1000 500 600 700 800

Wavelength [nm]900 1000

5000 0

5

10

15

0.2

0.4

0.6

0.8

1.0

1.2

600 700 800

Apple (GD)

PearKiwifruit

PeachApple (Fuji)

Wavelength [nm](a) (c)

(b) (d)

m a [c

m�

1 ]m a

[cm

�1 ]

m� s [c

m�

1 ]m s

[cm

�1 ]

900 1000 500 600 700 800

Wavelength [nm]900 1000

FIGURE 5.2.7 Optical properties measured for selected fruit and vegetable samples: (a) and(b) spectra of the absorption coefficient; and (c) and (d) spectra of the reduced scatteringcoefficient (GD, ‘Golden Delicious’; RD, ‘Red Delicious’).

due to its low amoun t of chlorophy ll con tent, as sugges ted by its white flesh. For thespectral range less than 550 nm, a bsorption due to carot enoids in the fruit tissue,which has one absorp tion peak at 480 nm (Mer zlyak et al. 2003), b ecame moreevident for ‘ Golden Delicious ’ and ‘ Fuji ’ a pples, the peach, the pear, and thekiwifruit . For wavelengt hs betw een 730 and 900 nm, the absorption values for allfruit a nd vegeta ble samples were relatively small and stable.

No disce rnabl e chlorophyll absorp tion peak at 675 nm was observ ed from the ma

spectra for the plum , cucum ber, and zucchini squash samp les (Figur e 5.2.7b). Astrong absorp tion peak was found at 535 nm for the plum due to absorp tion byanthocy anins in the frui t tissue (Merzlyak et al. 2003), and at the same spect ralposition , a smal ler absorp tion peak was also observ ed for the ‘ Red Del icious ’ app lefor the same reason . The cucumber and the zucchi ni squash show ed similar patternsfor their absorp tion spect ra, which, h owever, wer e conspi cuousl y different fromthose of o ther frui t samples. The ma spectra of these tw o vegeta ble samples peakedat 550 and 720 nm, whi ch were probably attributed to the absorp tion by turmeri c andother pigme nts in the vegeta ble tissue.

The reduced scatt ering spect ra of the fruit and vegeta ble samp les did not showparticula r spect ral featu res o ther than the trend of stead ily decreas ing values with theincrease of wavelengt h (Figure 5.2.7 c and d). The pe ach samp le had the highes t m 0svalues over the entire spect ral regio n from 500 to 10 00 nm, while the m0

s values werelower for the kiwifr uit and the plum. Ot her samp les had intermed iate v alues for thereduced scatt ering coef ficient .

Figure 5.2.8 shows the mean and � 1.0 stand ard deviation (SD) spect ra for theabsorption coef ficient from 600 ‘ Golden Delici ous ’ apples, whi ch were measuredafter they had been kept in con trolled atm osphere (CA) envir onmen t (at 0 8 C with2% O2 and 3% CO2) for about 5 months (unpublished data). There were great


0.0

0.2

0.4

0.6

0.8

1.0

1.2

500 600 700 800 900 1000

Wavelength [nm]

Abs

orpt

ion

coef

ficie

nt [c

m�

1 ]Mean±1.0 SD

FIGURE 5.2.8 Mean and �1.0 SD spectra for the absorption coefficient obtained from 600‘Golden Delicious’ apples.

variation s in the values of ma among the samp les. The varia tion of ma values amongthe apples was especiall y ev ident in the visible spect ral regio n. Similar ly, largevariation s in the m0

s spect ra were also observ ed for these apples over the wave lengthsof 500 –1000 nm.

5.2.4.2 Evaluation of Appl e Fruit Firmness

Si nc e firmness is related to the structural characteristics of a fruit, one can expect thatabsorption and scattering parameters would be useful for evaluating fruit firmness.Spectra of the 600 ‘Golden Delicious’ apples (Figure 5.2.8) were used to predict theMT fruit firmness. The firmness of each fruit was measured using the MT firmness testerright after the scattering images were acquired. Firmness prediction models relating ma

or m0s to the MT firmness were developed using partial least squares regression coupled

with cross validation for the 400 calibration samples. In addition, the combined data ofma and m0

s were used for predicting fruit firmness using a stepwise multi-linear regressionmethod. Model validation results for the 200 validation samples are shown in Figure5.2.9. Both ma and m0

s were correlated to fruit firmness, with r ¼ 0.83 and 0.70,respectively. Moreover, when ma and m 0s are combined, better predictions of fruitfirmness were achieved with r ¼ 0.86. These results indicate that both absorption andscattering properties are related to fruit firmness and better firmness predictions can beachieved with the combined data of ma and m0

s than using single optical parameters.Hence, measurement of the absorption and scattering properties can provide a newmeans for evaluating internal quality of fruit and other food products.

5.2.4.3 Estimatio n of L ight Pene tration Depths in Fruit

Knowled ge of ligh t distrib ution and penetr ation depth in a particula r type of frui t canprovide a guide in designing an effective optical sensing configuration for qualityevaluation. For example, this knowledge could be used to determine the optimal


r =0.8390

30

40

50

60

70

80

90

Actual MT firmness [N]

Pre

dict

ed M

T fi

rmne

ss [N

]

(c)

r =0.86SEP = 6.07

30 40 50 60 70 80 90

90

SEP = 6.25r =0.70

SEP = 8.82

30

40

50

60

70

80

30 40 50 60 70 80 90 30 40 50 60 70 80 90

Actual MT firmness [N]

Pre

dict

ed M

T fi

rmne

ss [N

]

(a)

30

40

50

60

70

80

Actual firmness MT [N]

Pre

dict

ed M

T fi

rmne

ss [N

]

(b)

FIGURE 5.2.9 Prediction of the measured MT fruit firmness for ‘Golden Delicious’ usingthe (a) absorption, (b) reduced scattering coefficients, and (c) their combined data.

distance between the light inci dent area and the detecting area, so that the lightsignals acquired would have reached the targe t tissue of the frui t to be meas ured.

Direct meas urement of light penetr ation depths in a fruit can be dif ficult and maynot be practical without d estroying the sample. Researche rs inves tigated light pene-tration depths in severa l types of fruits including manda rin and apple (Lamm ertynet al. 2000; Fraser e t al. 2002). The ligh t penetrati on depths report ed in the literaturediffer in their values due to different instrum ent setup s an d different de finitions used.With the knowledge of optical param eters, one can easily esti mate light penetr ationdepths without cond ucting expensive experi ments. For instance, based on diffu siontheory, the ligh t p enetratio n depth (d ) may be e stimated using the follow ing equation(Wilson and Jacques 1990):

d ¼ 1meff

¼ 1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi3ma ma þ m0

s

� �q ( 5: 2:21 )

Light pen etration depth is de fined as the distance requi red for the light inten sity levelto reduce to a factor of 1=e, or � 37% (Equatio n 5.2.21). Other resear chers de finedthe ligh t penetrati on depth to be at 1% relat ive trans missio n intensity in the fruittissue (Fraser et al. 2002). Since ma and m0

s are wavelengt h- and fruit-dep endent , ligh tpenetrati on dep ths will change with wave length and vary from frui t to fruit. Figure5.2.10 shows the mean light penetr ation depths for ‘Gol den Delici ous’ applesand � 1.0 SD using the 1% transmi ssion attenuati on (� 4.6 � d), based on the data


0

1

2

3

4

5

6

500 600 700 800 900 1000

Wavelength [nm]

Pen

etra

tion

dept

h [c

m]

FIGURE 5.2.10 Light penetration depths for 600 ‘Golden Delicious’ apples estimated fromthe mean values of the absorption and reduced scattering coefficients and their �1.0 SD usingEquation 5.2.21 (or 4.6� d at 1% transmittance).

for the absorp tion coef ficient presen ted in Figure 5.2.8 and the corres pondin greduced scatt ering coef ficient (not presen ted). The maxi mum light penetrati on de pthswere in the NIR regio n betwee n 700 and 900 nm. For a given wave length, there wasa large variation in the penetr ation depth among the apple samples. Fo r examp le, theestimated penetr ation depth at 800 n m using � 1.0 SD of the mean ma and m0

s rangedfrom 1.6 to 4.3 cm among the apples . The light penetr ation depth was greatlyaffected by the absorp tion of light by pigm ents and water in the fruit.

5.2.4.4 Monte Car lo Simul ation of Light Propag ation in Apple Fruit

While analyt ical solution s to the diffusion equati on may be obtai ned for turbid mediaof simp le geome tries and speci fic types of ligh t source (such as the one descri bed inFigure 5.2.1), many probl ems of pract ical interest can not be solved analytical ly. Forgeneral light propaga tion probl ems, we have to resort to numerical meth ods and theMonte Carlo (MC) metho d is the most widely used. Monte Car lo simu lation is astochastic numer ical approac h that employs random numbe rs in solvi ng the prob-lems of ligh t trans fer in turbid medi a. In applyi ng an MC approac h to photontransport , single photon s are trace d throu gh each step and the dist ribution of lightin the medi um is buil t based upon the trajector ies of individua l photons. Theparameter s o f each step are calcul ated using funct ions whos e arguments are randomnumbers. As the number of ph otons incre ases to infi nity, the MC predi ction for thelight transfer app roaches an exact solution of the d iffusion equation. Several MCsimulation progra ms are alrea dy publi cly av ailable for studying light trans fer inturbid media (e.g., MCML and CONV [Wang et al. 1995a,b] and tMCImg [Boaset al. 2002]).

MC simulations were performed for quantifying diffuse reflectance and photonabsorption inside an apple fruit illuminated by a cw point light source as described in


0

5

10

15

20

(a) (b)

(c) (d)

−20 −10 0

Radial distance [mm]

ma 1 and ms� 1

ma 2 and ms� 1 ma 2 and ms� 2

ma 1 and ms� 2

0.1

0.010.01

0.01

0.001

0.001

0.001

0.1

0.01

0.001

Dep

th [m

m]

10 20

0

5

10

15

20−20 −10 0


Dep

th [m

m]

10 20

0

5

10

15

20−20 −10 0


Dep

th [m

m]

10 20

0

5

10

15

20−20 −10 0


Dep

th [m

m]

10 20

FIGURE 5.2.11 Light distribution patterns in apple fruit estimated by Monte Carlo simula-tions. The absorption (ma) and reduced scattering (m0

s) coefficients used for the simulationswere ma,1¼ 0.39 cm�1, ma,2¼ 0.04 cm�1; m0

s,1 ¼ 21.45 cm�1, m0s,2 ¼ 8.63 cm�1. (From Qin, J.

and Lu, R., ASABE Paper No. 073058, American Society of Agricultural and BiologicalEngineers, St. Joseph, MI, 2007b.)

Figure 5.2.2 (Qin and Lu 2007b). Four pairs of ma and m0s values wer e chosen for MC

simulations to demon stra te how the chang e of ma and m0s values would change the

pattern of ligh t distrib utions in the fruit.Figure 5.2.11 shows light absorp tion powe r densi ties as contou r maps ve rsus

sample dep th and radia l distance inside a ‘ Golden Del icious ’ apple with each of thefour pairs of ma and m0

s values . The pattern of a bsorption was greatly shaped by thecombinati ons of ma and m0

s values. The fruit tissue with a large ma value absorb ed ligh tenergy rapid ly in a shorter distance, making it more dif ficult for the ligh t to propaga teinto the deeper and broader areas (Figure 5.2.11a and b). When the ma value isconstant, the tissue with a larger m 0s value would prevent light from penetr ating thedeeper layer s of the fruit tis sue (Figur e 5.2.11a versus b and Figure 5. 2.11c versus d).In summ ary, a fruit with small values of ma and m0

s woul d hav e smal ler attenuati on or alarger light penetr ation depth, whereas a fruit wi th larger values of ma and m0

s wouldhave great er attenuati on or a smaller light penetr ation depth. Light penetr ation depth isdetermin ed by the combi nation of ma and m 0s values for the fruit.

Monte Carlo simulat ion can help to estimate the effective detect ing distance for aspeci fic sensing con fi guration, i.e., when the speci fic light source and the detector arealready selec ted. MC simulation resul ts are also useful in determin ing an opti malsensing con figurat ion if the samp le condit ions and detect ion distance are alrea dyknown (Qin and Lu 2007b).


5.2.5 LIGHT -SCATTERING T ECHNIQUE F EASIBLE FOR ASSESSING F RUIT

FIRMNESS IN P RACTICE

The spati ally resol ved hypers pectr al imaging techni que descri bed in Secti on 5.2.3allows us to measure and separate the absorption and scattering parameters. Theoptical property data are useful for quantitative analysis of light propagation anddistribution in the sample and estimation of light penetration depths. Moreover, theoptical property data can be used for assessing quality attributes such as firmness andsoluble solids content. However, this fundamental approach requires sophisticatedalgorithms that may be susceptible to fitting error in extracting the values of ma andm0s for each wavelength. The technique is not suitable for rapid assessment of the

texture of individual product items, which is required in product sorting and grading.This section presents a practical light-scattering technique for acquiring spectralscattering images from individual fruit and the mathematical models to quantifyspectral scattering characteristics for prediction of fruit firmness.

The concept of utilizing light-scattering characteristics to assess product qualitywas first proposed by Birth and colleagues in the 1970s (Birth 1978, 1982; Birthet al. 1978). A transmission technique with a laser at 632 nm as a light source wasused to measure the scattering and absorption parameters of the Kubelka–Munkmodel. Birth and coworkers (1978) reported that the scattering coefficient could beused for evaluating pork quality. However, the method and technique proposed wereinconvenient and rather primitive by today’s standard. Little progress was made onlight-scattering research in the 1980s. Interest in the technique was renewed in the1990s, and several studies were reported on the measurement of fruit quality=matur-ity by analyzing the reflectance images or profiles from an intact fruit generated by avisible or NIR laser. Tu and coworkers (2000) performed a study of using a He–Nelaser at 670 nm to generate scattering images at the surface of tomatoes to assess theirmaturity levels. They reported that the total number of pixels recorded by the redband CCD above a specific threshold value was related to the maturity levels of thetomatoes. McGlone and associates (1997) studied the intensities of scattered lightfrom kiwifruit, which was generated by a diode laser at 864 nm, in relation to fruitfirmness. They reported that the intensities of the scattered light from the fruitincreased with decreasing firmness, especially at large distances. However, sincefirmness is such a complex phenomenon, light scattering at single wavelengthscannot provide sufficient information about the structural characteristics of thefruit, and thus is unable to provide accurate measurement of fruit firmness.

Lu (2004) introduced the concept of multispectral scattering for assessing thequality attributes, i.e., firmness and soluble solids content, of apples. The conceptwas based on the premise that scattering at multiple wavelengths would providemore information about the structural characteristics of a fruit sample and thus couldlead to better assessment of fruit firmness. A broadband light source was used togenerate scattering images at the surface of apple fruit. Scattering images wereacquired at five wavelengths using a high-performance CCD camera coupled to amechanical filter wheel installed with five band-pass filters. One-dimensional scat-tering profiles were obtained from the 2-D scattering images based on their sym-metry to the light incident point. Individual scattering profiles were input into a


neural network for predicting the firmness and soluble solids content of apples.Relatively good corrections were obtained for both firmness and soluble solidscontent. In subsequent studies, Peng and Lu (2005, 2006b,c, 2007) proposedimproved hardware designs and mathematical models for measuring and analyzingspectral scattering images and obtained good predictions of fruit firmness.

To effectively predict fruit firmness and other quality attributes with the multi-spectral scattering technique, it is important to select appropriate wavelengths anduse proper mathematical models, which are described in the following subsections.

5.2.5.1 Wavelengths Selection

Selection of appropriate wavelengths is the first step in implementing the multispectralscattering technique for assessing fruit firmness and other quality attributes suchas soluble solids content. Ideally, the number of wavelengths selected should besmall, e.g., not more than four to five, and they can provide maximum informationabout the structural and chemical properties of a particular type or variety of fruit.Considerable works in NIR spectroscopy have been reported on measurement of thefirmness and solid content of fruits, and they can provide some guide inthe preliminary selection of wavelengths for the multispectral scattering method.Because of the complexity of light–tissue interactions and large variations in proper-ties and characteristics for different varieties, different sets of wavelengths have beenreported. For example, Lu (2004) selected five wavelengths (680, 880, 905, 940, and1060 nm) in his study on apples. The wavelength of 680 nm was related to theabsorption of light by chlorophyll in the fruit. Chlorophyll content is an importantindicator of fruit maturity and thus has an important implication on fruit firmness.Light at the wavelengths of 880 and 905 nm has better penetration into the fruit tissueand hence would be helpful for assessment of the properties of fruit flesh. Thesewavelengths are also useful for soluble solids measurement. The wavelengths of 940and 1060 nm are useful for firmness and soluble solids prediction. Qing et al. (2007)also used a set of five wavelengths (680, 780, 880, 940, and 980 nm) in their study ofassessing apple fruit quality.

A better, but more expensive and time-consuming, approach to wavelengthsselection is to conduct a complete wavelength search by acquiring spectral scatteringimages at individual wavelengths over a spectral range of interest. Peng and Lu(2006b,c) acquired 36 scattering images between 650 and 1000 nm for ‘GoldenDelicious’ and ‘Red Delicious’ apples using a monochromatic CCD camera coupledwith an LCTF. By applying stepwise multi-linear regression analysis coupled withcross validation, they found a set of optimal wavelengths. The top five wavelengthsfor firmness prediction were in the spectral range between 690 and 990 nm.

5.2.5.2 Instrumentation

After the wavelengths are determined, one needs to choose an instrumentation setupfor acquiring spectral scattering images. The instrumentation setup for scatteringmeasurement will depend on application needs and research goals. A typical spectralscattering system consists of an imaging device with a computer, a band-pass filter,and a light source. Both color and monochromatic CCD cameras have been used.


Since light attenuates rapidly in the fruit tissue, the intensities of reflectance imageswill decrease drastically in a short distance from the beam incident point. Hence, it isimportant that the CCD imager have a large dynamic range. The imager should alsohave high photon efficiencies especially in the short-wave NIR range (700–1000nm), which is needed for scattering measurement. Monochromatic CCD imagers arepreferred because they have larger dynamic ranges, e.g., 12 or 16 bits versus 8 bitsfor the color CCD detectors.

Selection of a light source is critical for the measurement of light scattering. Twotypes of light sources are currently in use: monochromatic or laser light andbroadband light. To better quantify the scattering characteristics, the light beamshould be small in size (e.g., 2 mm or smaller). Laser is an excellent monochromaticlight source (Lu and Peng 2007; Qing et al. 2007); it can deliver more light power perunit area for a given wavelength than a broadband light source, which is needed forfast acquisition of scattering images. However, to acquire scattering images atmultiple wavelengths, one will need multiple lasers, which can be expensive andinconvenient. A broadband light source can provide all the needed wavelengths (Lu2004; Peng and Lu 2006b). The main drawback of the broadband light source is itslow output power per unit area because considerable light would be lost when thelight from the lamp is coupled to a single optic fiber cable.

Figure 5.2.12 shows a multispectral imaging system equipped with an LCTF foracquiring spectral scattering images between 650 and 1000 nm. The system consistsof a multispectral imager and a broadband light source that is coupled to an opticfiber with a collimating lens to generate a focused light beam (�1.5 mm in size). TheLCTF is electronically tunable for rapid selection of any specific wavelength. Sinceindividual spectral scattering images are acquired one at a time, the system shown in

Computer

Controller

CCD camera

Optic fiber

Light intensitycontroller

Lens

Fruit

Focusing lens

Control cables

Video capture card

Sync cardLCTFIncidencelight beam

Scattering image

Lamp

Lightsensor

Light sourcesupplier

FIGURE 5.2.12 Schematic of a multispectral imaging system installed with an LCTF foracquiring spectral scattering images from apple fruit. (From Peng, Y. and Lu, R., PostharvestBiol. Technol., 41, 266, 2006d.)


Figure 5.2.12 is not suitable for real-time or onli ne sorting and grading appli cations.Lu and Peng (2007) developed a laboratory laser-based multispectral imagingprototype for real-time assessment of apple firmness. A common aperture CCDcamera was coupled with a multispectral imaging spectrograph for simultaneousacquisition of spectral scattering images at four selected wavelengths. A custom-builtmulti-laser unit was used for the prototype, which had four lasers (680, 880, 905, and940 nm) that were coupled into a single optic fiber with a micro-lens to generate onesingle beam of 1 mm size. The output power of the four lasers could be adjustedindividually to achieve desirable scattering images at the four wavelengths. Theexposure time for the scattering image acquisition could be as short as 10 ms, whichwould meet the requirement for online sorting and grading applications.

5.2.5.3 Mathematical Description of Light-Scattering Profiles

After scattering images have been acquired, it is critical that appropriate mathema-tical methods and procedures be used to describe the scattering characteristics.Figure 5.2.13 shows a typical scattering image acquired from an apple fruit andthe method=procedure of processing and analyzing the scattering image. Preproces-sing of the scattering images may be needed to remove noisy signals or pixels and

Scatteringimage is dividedinto N circularbands of equal

pixels (or distance)

300Maximum peak value

of light intensity

Peak area

I = a +b

xc

d

Full scatteringwidth at half

maximum (FWHM)

Scattering gradientin descending corner

Scattering area

b

b

2

2c

d

250

200

Ligh

t int

ensi

ty

150

100

50

0−150 −100 −50 0

Asymptotic valueof light intensity

Distance [pixels]a

50 100 150

Saturation area

Scattering area 1+

12 3

N...

(a)

(b) (d)

(c)

FIGURE 5.2.13 Spectral scattering images of an apple fruit in (a) 2-D display format and(b) 3-D display format; (c) the radial averaging of scattering image pixels; and (d) the resultant1-D scattering profile fitted by the modified Lorentzian distribution function. (From Peng, Y.and Lu, R., Trans. ASABE, 49, 259, 2006b.)


http://www.crcnetbase.com/action/showImage?doi=10.1201/9781420054033.ch5&iName=master.img-012.jpg&w=138&h=96

improve image quality. Fo r instance, many apples hav e small, isolated spots on theirskin, whos e pigm ents are distinct ly diff erent from the rest of the fruit. The presen ceof these spots at the fruit surface could ca use unusual ly low (dark) or high (brig ht)refl ectanc e on the 2-D scatteri ng ima ge, which in turn woul d affect the resul tant 1-Dscattering pro files. Peng and Lu (2006d) sugges ted a filtering method to remo vethese abnorm al pixel s from the scatteri ng ima ges to improve the ima ge quality.

Scattering ima ges acquir ed from a frui t are general ly symmet ric to the beamincident center . Henc e after the prepro cessi ng of the scatteri ng image s, we can utilizethe symmet ry featu re to reduce each 2-D scatteri ng ima ge to a 1-D scatteri ng pro file.This is ach ieved by first dividing the scatt ering images into a numbe r of concent ricrings of equal distance and then performing the radial averaging of all pixels withineach circu lar ring (Figure 5.2.13c) . The resul tant 1-D scatt ering pro file (Figur e5.2.13d) consists of two sections: a saturation section and an unsaturated, scatteringsection. Both saturated and unsaturated sections contain information about thescattering characteristics of a fruit and hence are useful for firmness assessment.Before further analysis of the scattering profiles, it is recommended that eachscattering profile be corrected for the effect of fruit size using Equation 5.2.19 orusing a simpler equation proposed by Peng and Lu (2006d).

Different mathematical functions may be used to describe the scattering profilesof fruit. The modified Lorentzian distribution function in Equation 5.2.22 and themodified Gompertz function in Equation 5.2.23 are two functions that were foundto be appropriate for characterizing the scattering profiles of fruit (Peng and Lu2006a,b,d, 2007).

RL(r) ¼ aþ b

1þ rc

d (5:2:22)

and

RG(r) ¼ aþ b 1� e� exp («�dr) �

(5:2:23)

wherea, b, c, and d are Lorentzian parametersa, b, «, and d are Gompertz function parameters

Lorentzian functions are commonly used for describing light or signal distributionpatterns in the optical and electric research fields, while Gompertz functions havebeen used for describing animal and organism growth (Peng and Lu 2007). Eachparameter in the above equations represents certain characteristics of the scatteringprofile. In the modified Lorentzian form of Equation 5.2.22, the parameter a repre-sents an asymptotic value, b is the peak value, c is related to the scattering width, andd reflects the slope of the scattering profile. Similarly, each parameter in Equation5.2.23 is associated with a specific characteristic of the scattering profile. Examin-ation and comparison of these parameters and their contributions to shaping thescattering profiles and firmness prediction is presented in Peng and Lu (2007).


Simpler funct ion forms o r varia nts with three or tw o param eters can be deriv ed fromthe four- param eter Lorentzi an distributio n function and the four- param eter Gom pertzfunction.

After a mathemat ical funct ion is selec ted, we can u se a nonlinear curve- fittingalgorithm to fit the function to each scattering pro file, from which values of indi vi-dual function parameter s are obtai ned. Compar ison of different funct ions showedthat the four- parameter Lorentzian funct ion (Equati on 5.2.22) and the four- param eterGompert z function (E quation 5.2.23) could better de scribe the scatt ering pro fi lesthan thei r sim pler funct ion varia nts with two or three parameter s and other funct ionssuch as Gaussia n funct ion. The modi fied Gom pertz funct ion (Equati on 5.2.23) wasslightly bett er in the predictio n of fruit firmne ss than the modi fied Lorentzi anfunction, but the form er had slower converg ence rates in the curve- fitting proces s(Peng and Lu 20 07).

Other methods of analyz ing the scatteri ng pro files have also been propos ed. Onemethod was to use the en tire scatteri ng pro files as inputs to a neural network syste mfor predi cting fruit firmne ss (Lu 2004). Instead of perfor ming radial averagi ng, thehistogram s of pixels for the scatteri ng ima ges can be calculated above a giventhreshold value and used to descri be the scatteri ng characteri stics (Qing et al.2007). This pixel –hist ogram calculati on ap proach is sim pler and fast er compa redwith the mathemat ical modeling approac h described above.

5.2.5.4 Fruit Firmn ess Assess ment

Scattering param eters or scatteri ng ima ge featu res are useful for predi cting fruitfirmness and other quali ty attribut es. Depending on the met hods used to proces sand analyz e the scatteri ng ima ges=pro files, different model predic tion approac hesmay be used. A sim ple mul tilinear regres sion model can be establis hed relat ing thescatteri ng funct ion parame ters for each wave length to the firmne ss of apples . Figure5.2.14 show s resul ts on using Lorentzi an param eters (Equati on 5.2.22) at sevenwavelengt hs for predictin g the MT firmne ss of ‘ Golden Delici ous ’ an d ‘RedDelicious ’ apples for the cali bration and vali dation sets , respec tively. The a ppleswere harves ted in 2 004, and they had been kept either in refrigerat ed air or CAenvironmen t for severa l mont hs before the testing. Scatt ering images were acq uiredat eight wave lengths for ‘Gol den Delicio us’ (650, 680, 700, 74 0, 820, 880, 910, and990 nm) and at seven wavelengt hs for ‘ Red Delici ous ’ (680, 700, 740, 800, 820, 910,and 990 nm). The se wave lengths were consid ered opti mal for each variety based onthe results from a previous study (Peng and Lu 2006c). The scattering images wereobtained using a low resolution, 8-bit monochromatic CCD camera equipped with anLCTF dev ice (Figure 5.2.12). The filter ing method discussed earlier was used toremove low- and high-noise pixels and the scattering profiles were corrected for thefruit size effect using a simplified correction equation (Peng and Lu 2006d). Theresults presented in Figure 5.2.14 are quite remarkable considering the fact that a lowresolution (512� 512 pixel), 8-bit CCD camera was used in the study. Peng and Lu(2006d) compared the firmness prediction results from ‘Golden Delicious’ and ‘RedDelicious’ apples from multispectral scattering with those from NIR measurements


r = 0.899SEC = 6.13

20

30

40

50

60

70

80

90

MT firmness [N]

Est

imat

efir

mne

ss[N

]r = 0.897

SEP = 6.14

20

30

40

50

60

70

80

90

20

MT firmness [N]

Pre

dict

edfir

mne

ss[N

]P

redi

cted

firm

ness

[N]

r = 0.908SEC = 6.36

20

30

40

50

60

70

80

90

100

MT firmness [N]

Est

imat

efir

mne

ss[N

]

r = 0.898SEP = 6.41

20

30

40

50

60

70

80

90

100

30 40 50 60 70 80 9020 30 40 50 60 70 80 90

20 30 40 50 60 70 80 90 100

MT firmness [N]20 30 40 50 60 70 80 90 100

FIGURE 5.2.14 Calibration (left) and validation (right) results on prediction of the MT fruitfirmness for (a) ‘Golden Delicious’ and (b) ‘Red Delicious’ apples using the four parameters ofLorentzian distribution function (Equation 5.2.22) for the scattering profiles (SEC, standarderror for calibration; SEP, standard error for prediction or validation). (From Peng, Y. and Lu,R., Postharvest Biol. Technol., 41, 266, 2006d.)

over the spectral region of 500–1000 nm. Even with no corrections for fruit size andnoise pixels, multispectral scattering had better firmness predictions (r¼ 0.82 and0.81 for ‘Red Delicious’ and ‘Golden Delicious’, respectively) than NIR spectros-copy (r¼ 0.50 and 0.48). Using the histogram of scattering image pixels to charac-terize the scattering features, good firmness prediction results for ‘Elstar’ and‘Pinova’ apples were obtained with r¼ 0.90 (Qing et al. 2007).

5.2.6 CONCLUSIONS AND NEEDS FOR FUTURE RESEARCH

The spatially resolved hyperspectral imaging technique described in this sectionprovides a new way of measuring the absorption and scattering properties ofhorticultural products. The optical parameters determined from the fruit and vege-table samples allow us to analyze light propagation and distributions in theseproducts under different lighting and sensing configurations. They are also usefulfor predicting a quality attribute such as fruit firmness. Compared with time-resolvedand frequency-domain techniques, the spatially resolve technique is simpler, easierto use, and more suitable for horticultural and food products. The technique is,however, based on a diffusion model that is derived for homogeneous semi-infiniteturbid media, and thus it has limitations. Many horticultural products have a surface


layer (skin) whose optical properties are quite different from those of the interiorlayers. Hence, care should be taken in interpreting the optical properties measured bythe spatially resolved technique since they actually reflect both the skin and flesh.Further research should be performed to answer the question of how much fruit skinwould contribute to the measured optical properties. Moreover, a diffuse theorymodel for two-layered or multilayered scattering media should be considered tobetter measure the properties of interior tissue of horticultural products. This willinevitably present more challenges than the simple diffusion model described in thissection, but could lead to more accurate characterization of the optical properties ofthe interior layers of horticultural products.

For practical applications, light scattering at multiple wavelengths has showngreat potential for assessing the texture or firmness of horticultural products. Thisemerging technique provides better characterization of light scattering in suchhorticultural products as apples than does conventional NIR spectroscopy. Thelight-scattering method is relatively simple and fast, and can be potentially imple-mented for online sorting and grading of fruit for firmness. There are, however, afew issues that should be considered in further research. First, improved instru-mentation designs are needed so that the imaging system can acquire scatteringimages at multiple wavelengths simultaneously and rapidly, which is critical forproduct sorting and grading. Second, the shape of a fruit will influence themeasurement of actual reflectance intensities from the sample. Although the fruitsize correction equations have been developed, they are only suitable for the fruitof spherical shape. A better size=shape correction method is needed when productitems to be inspected are of complicated geometry or deviate significantly from theassumed shape. Third, like those obtained with the hyperspectral imaging tech-nique, the scattering profiles measured with the light-scattering technique are alsoaffected by the surface layer of the sample. An appropriate method should bedeveloped for minimizing the effect of fruit skin on the measurement of light-scattering profiles. Finally, the light-scattering technique, similar to NIR spectros-copy, relies on the establishment of appropriate calibration models to achieveaccurate predictions of fruit firmness. The transferability of the calibration modelsfrom one instrument to another is an issue that has not been addressed so far. Theprediction models developed in reported studies were usually validated against thesamples that had the pre- and postharvest histories that were identical or similar tothe calibration samples. The robustness of the calibration model for predictingsamples from a new population (influence of season, location, climate, etc.) needsto be investigated and an effective method of updating the model should beconsidered. With proper resolution of these issues, the light-scattering techniquecan become an important tool for nondestructive assessment of postharvest quality ofhorticultural products.

ACKNOWLEDGMENT

The author wishes to thank Dr. Jianwei Qin and Dr. Yankun Peng, who wereformerly with the Department of Biosystems and Agricultural Engineering atMichigan State University, East Lansing, Michigan, for their assistance in thepreparation of this section.


REFERENCES

Abbott, J.A., R. Lu, B.L. Upchurch, and R.L. Stroshine. 1997. Technologies for nondestruc-tive quality evaluation of fruits and vegetables. In: Horticultural Reviews 20, eds.J. Janick, John Wiley & Sons Inc., ISBN: 0-471-18906-5, pp. 1–120.

Birth, G.S. 1978. The light scattering properties of foods. Journal of Food Science 16:916–925.Birth, G.S., C.E. Davis, and W.E. Townsend. 1978. The scattering coefficient as a measure of

pork quality. Journal of Animal Science 46:639–645.Birth, G.S. 1982. Diffuse thickness as a measure of light scattering. Applied Spectroscopy

36:675–682.Boas, D.A., J.P. Culver, J.J. Stott, and A.K. Dunn. 2002. Three dimensional Monte Carlo code

for photon migration through complex heterogeneous media including the adult humanhead. Optics Express 10:159–170.

Bourne, M.C. 2002. Food Texture and Viscosity, 2nd Edn., Academic Press, San Diego, USA.ISBN: 0-12-11062-5.

Cubeddu, R., C. D’Andrea, A. Pifferi et al. 2001. Nondestructive quantification of chemicaland physical properties of fruits by time-resolved reflectance spectroscopy in thewavelength range 650–1000 nm. Applied Optics 40:538–543.

Dam, J.S., P.E. Andersen, T. Dalgaard, and P.E. Fabricius. 1998. Determination of tissueoptical properties from diffuse reflectance profiles by multivariate calibration. AppliedOptics 37:772–778.

Farrell, T.J., M.S. Patterson, and B. Wilson. 1992. A diffusion-theory model of spatiallyresolved, steady-state diffuse reflectance for the noninvasive determination of tissueoptical-properties in vivo. Medical Physics 19:879–888.

Fraser, D.G., R.B. Jordan, R. Künnemeyer, and V.A. McGlone. 2002. Light distribution insidemandarin fruit during internal quality assessment by NIR spectroscopy. PostharvestBiology and Technology 27:185–196.

Gobin, L., L. Blanchot, and H. Saint-Jalmes. 1999. Integrating the digitized backscatteredimage to measure absorption and reduced-scattering coefficient in vivo. Applied Optics38:4217–4227.

Groenhuis, R.A.J., H.A. Ferwerda, and J.J. Tenbosch. 1983. Scattering and absorption ofturbid materials determined from reflection measurements–1. Theory. Applied Optics22:2456–2462.

Harker, F.R., J.H. Maindonald, and P.J. Jackson. 1996. Penetrometer measurement of apple andkiwifruit firmness: Operator and instrument differences. Journal of ASHS 121:927–936.

Kienle, A., L. Lilge, M.S. Patterson, R. Hibst, R. Steiner, and B.C. Wilson. 1996. Spatiallyresolved absolute diffuse reflectance measurements for noninvasive determination of theoptical scattering and absorption coefficients of biological tissue. Applied Optics35:2304–2314.

Kortüm, G. 1969. Reflectance Spectroscopy: Principles, Methods, Applications. Springer-Verlag, LCCCN: 79-86181.

Lammertyn, J., B. Nicolaï, K. Ooms, V. De Smedt, and J. De Baerdemaeker. 1998. Non-destructive measurement of acidity, soluble solids, and firmness of Jonagold applesusing NIR-spectroscopy. Transactions of the ASAE 41:1089–1094.

Lammertyn, J., A. Peirs, J. De Baerdemaeker, and B. Nicolaï. 2000. Light penetrationproperties of NIR radiation in fruit with respect to non-destructive quality assessment.Postharvest Biology and Technology 18:121–132.

Lawrence, K.C., B. Park, W.R. Windham, and C. Mao. 2003. Calibration of a pushbroomhyperspectral imaging system for agricultural inspection. Transactions of the ASAE46:513–521.

Lu, R. and Y.R. Chen. 1998. Hyperspectral imaging for safety inspection of food andagricultural products. SPIE Proceedings 3544: 121–133.

Lu, R., D.E. Guyer, and R.M. Beaudry. 2000. Determination of firmness and sugar content ofapples using near-infrared diffuse reflectance. Journal of Texture Studies 31:615–630.


Lu, R. 2001. Predicting firmness and sugar content of sweet cherries using near-infrareddiffuse reflectance spectroscopy. Transactions of the ASAE 44:1265–1271.

Lu, R. 2003. Detection of bruises on apples using near-infrared hyperspectral imaging.Transactions of the ASAE 46:523–530.

Lu, R. 2004. Multispectral imaging for predicting firmness and soluble solids content of applefruit. Postharvest Biology and Technology 31:147–157.

Lu, R. and J.A. Abbott. 2004. Force=deformation techniques for measuring texture. In: Texturein Food: Volume 2: Solid Foods, Ed. D. Kilcast, Woodhead Publishing Limited, ISBN:1-85573-724-8, pp. 109–145.

Lu, R. and Y. Peng. 2007. Development of a multispectral imaging prototype for real-timedetection of apple fruit firmness. Optical Engineering 46(12), 123201.

Martinsen, P. and P. Schaare. 1998. Measuring soluble solids distribution in kiwifruit usingnear-infrared imaging spectroscopy. Postharvest Biology and Technology 14:271–281.

McGlone, V.A., H. Abe, and S. Kawano. 1997. Kiwifruit firmness by near infrared lightscattering. Journal of Near Infrared Spectroscopy 5:83–89.

McGlone, V.A. and S. Kawano. 1998. Firmness, dry-matter and soluble-solids assessment ofpostharvest kiwifruit by NIR spectroscopy. Postharvest Biology and Technology13:131–141.

Merzlyak, M.N., A.E. Solovchenko, and A.A. Gitelson. 2003. Reflectance spectral featuresand non-destructive estimation of chlorophyll, carotenoid and anthocyanin content inapple fruit. Postharvest Biology and Technology 27:197–211.

Mourant, J.R., T. Fuselier, J. Boyer, T.M. Johnson, and I.J. Bigio. 1997. Predictions andmeasurement of scattering and absorption over broad wavelength ranges in tissuephantoms. Applied Optics 36:949–957.

Nichols, M.G., E.L. Hull, and T.H. Foster. 1997. Design and testing of a white-light, steady-state diffuse reflectance spectrometer for determination of optical properties of highlyscattering systems. Applied Optics 36:93–104.

Park, B., K.C. Lawrence, W.R. Windham, and R.J. Buhr. 2002. Hyperspectral imaging fordetecting fecal and ingesta contaminants on poultry carcasses. Transactions of the ASAE45:2017–2026.

Peng, Y. and R. Lu. 2005. Modelling multispectral scattering profiles for prediction of applefruit firmness. Transactions of the ASAE 48:235–242.

Peng, Y. and R. Lu. 2006a. Improving apple fruit firmness predictions by effective correctionof multispectral scattering images. Postharvest Biology and Technology 41:266–274.

Peng, Y. and R. Lu. 2006b. An LCTF-based multispectral imaging system for estimation ofapple fruit firmness: Part I. Acquisition and characterization of scattering images.Transactions of the ASABE 49:259–267.

Peng, Y. and R. Lu. 2006c. An LCTF-based multispectral imaging system for estimationof apple fruit firmness: Part II. Selection of optimal wavelengths and development ofprediction models. Transactions of the ASABE 49:269–275.

Peng, Y. and R. Lu. 2006d. Improving apple fruit firmness predictions by effective correctionof multispectral scattering images. Postharvest Biology and Technology 41:266–274.

Peng, Y. and R. Lu. 2007. Prediction of apple fruit firmness and soluble solids content usingcharacteristics of multispectral scattering images. Journal of Food Engineering82:142–152.

Qin, J. and R. Lu. 2006. Measurement of the optical properties of apples using hyperspectraldiffuse reflectance imaging. ASABE Paper No. 063037, American Society ofAgricultural and Biological Engineers, St. Joseph, MI.

Qin, J. 2007. Measurement of the optical properties of horticultural and food products byhyperspectral imaging. Ph.D. dissertation, Michigan State University, East Lansing, MI.

Qin, J. and R. Lu. 2007a. Measurement of the absorption and scattering properties of turbidliquid foods using hyperspectral imaging. Applied Spectroscopy 61:388–396.

Qin, J. and R. Lu. 2007b. Monte Carlo simulation of light propagation in apples. ASABE PaperNo. 073058, American Society of Agricultural and Biological Engineers, St. Joseph, MI.


Qing, Z., B. Ji, and M. Zude. 2007. Predicting soluble solid content and firmness in apple fruitby means of laser light backscattering image analysis. Journal of Food Engineering82:58–67.

Schaare, P.N. and D.G. Fraser. 2000. Comparison of reflectance, interactance, and transmit-tance modes of visible-near-infrared spectroscopy for measuring internal properties ofkiwifruit. Postharvest Biology and Technology 20:175–184.

Shmulevich, I. and M.S. Howarth. 2003. Non-destructive dynamic testing of apples forfirmness evaluation. Postharvest Biology and Technology 29:287–299.

Tu, K., P. Jancsok, B. Nicolaï, and J. De Baerdemaeker. 2000. Use of laser-scattering imagingto study tomato fruit quality in relation to acoustic and compression measurements.International Journal of Food Science and Technology 35:503–510.

Tuchin, V. 2000. Tissue Optics: Light Scattering Methods and Instruments for MedicalDiagnosis. SPIE Press, Bellingham, WA, USA. ISBN: 9780819434593.

Vo-Dinh, T. 2003. Biomedical Photonics Handbook. CRC Press, Boca Raton, FL, USA.ISBN: 0-8493-1116-0.

Wang, L.H., S.L. Jacques, and L.Q. Zheng. 1995a. MCML—Monte-Carlo modeling of lighttransport in multilayered tissues. Computer Methods and Programs in Biomedicine47:131–146.

Wang, L.H., S.L. Jacques, and L.Q. Zheng. 1995b. CONV—convolution for responses to afinite diameter photon beam incident on multi-layered tissues. Computer Methods andPrograms in Biomedicine 53:141–150.

Williams, P. and K. Norris. 2001. Near-Infrared Technology in the Agricultural and FoodIndustries, 2nd Edn., AACC, St. Paul, MN, USA. ISBN: 1-891127-24-1.

Wilson, B.C. and S.L. Jacques. 1990. Optical reflectance and transmittance of tissues:principles and applications. IEEE Journal of Quantum Electronics 26:2186–2199.

Xia, J., E.P. Berg, J.W. Lee, and G. Yao. 2007. Characterizing beef muscles with opticalscattering and absorption coefficients in VIS-NIR region. Meat Science 75:78–83.

5.3 NMR FOR INTERNAL QUALITY EVALUATIONIN HORTICULTURAL PRODUCTS

NATALIA HERNÁNDEZ SÁNCHEZ, PILAR BARREIRO ELORZA,AND JESÚS RUIZ-CABELLO OSUNA

This section aims at summarizing the applicability of NMR in the context of internalquality assessment in fruits and vegetables. It has been structured in five sections;some of them are legible for both skilled and unskilled readers, while others may beharder in the first attempt.

It has been written in such a way as to make sections reading independent.The overview and applications section is clearly a matter of interest for a generalreader, while the NMR basics and MRI fundamentals may refer to physical andmathematical concepts that will slow down the reading. Unfamiliar readers shouldfeel free to skip those sections in the first approach.

The need for overcoming a large amount of hindrances for a competent industrialapplication reinforces the inclusion of such physical and mathematical concepts. It isthe intention of the authors to provide a wide discussion of some technical detailsthat may have an important role on the success for the practical use of NMR and MRIin fresh fruits and vegetables, whether this goal is achieved or not is something to bestated by the reader.


5.3.1 OVERVIEW ON APPLICATIONS IN FRUITS AND VEGETABLES

Nuclear magnetic resonance spectroscopy (MRS) and magnetic resonance relaxo-metry (MRR) together with MRI have been explored since the 1980s to eva-luate their applicability to the inspection of internal quality aspects in fruits andvegetables.

NMR is an especially useful monitoring technique since the signal emitted froma sample is sensitive to the density of certain nuclei, chemical structure, molecular oratomic diffusion coefficients, reaction rates, chemical exchange, and other pheno-mena (McCarthy 1994), which give an enormous scope for applications.

The aqueous protons (1H) in soft tissues, as well as 13C, 15N, 19F, 23Na, and 31Pnuclei can be detected, although 1H shows by far the highest applicability to fruitand vegetables (Clark et al. 1997). Correlation between quality parameters or dis-orders and NMR measurements has been found for many products, as reviewedby Hills and Clark (2003) and Butz et al. (2005), and as pointed out in recentpublications: Létal et al. 2003; Hernández-Sánchez et al. 2004, 2006, 2007;Thybo et al. 2004; Chayaprasert and Stroshire 2005; Gambhir et al. 2005; Hernándezet al. 2005; Marigheto et al. 2005; Raffo et al. 2005; Brescia et al. 2007;Goñi et al. 2007; Tu et al. 2007. Table 5.3.1 summarizes the NMR applicationsaimed at addressing the correlation between quality factors or disorders and NMRmeasurements in fruits and vegetables.

TABLE 5.3.1Summary of Quality Attributes and Disorders in Fruits and VegetablesStudied by MRR, MRS, and MRI Techniques (Including PD Maps)

Product

Maturity=Sugar

Content

Bruises=Voids=Seeds

TissueBreakdown

HeatInjury

Chill=FreezeInjury Infections

FruitApple MRR=MRI MRI MRI=MRR

PD mapsAvocado MRS=MRIBanana MRR

Cherimoya MRR=MRIDurian MRS=MRI MRIKiwifruit MRR MRI=MRR MRIMandarin MRR PD maps MRI

Mango MRS=MRR MRI MRIMangosteen MRI MRIMelon MRS MRI MRI

Nectarine MRI MRR PD mapsOrange MRS=MRR MRI MRIPapaya MRR

Peach MRI MRR=MRI


TABLE 5.3.1 (continued)Summary of Quality Attributes and Disorders in Fruits and VegetablesStudied by MRR, MRS, and MRI Techniques (Including PD Maps)

Product

Maturity=Sugar

Content

Bruises=Voids=Seeds

TissueBreakdown

HeatInjury

Chill=FreezeInjury Infections

Pear MRI MRR=MRIPersimmon MRIPineapple MRR=MRI

Tangerine MRI MRI MRIWatermelon MRS MRI

BerriesBlueberry MRI MRIGrape MRS=MRRStrawberry MRI

Small, stone fruits (drupes)Cherry MRR=MRS MRIOlive MRI MRIPlum=prunes MRS

VegetablesCourgette=Zucchini MRICucumber MRI MRI

Onion MRIPotato MRR=MRI MRI MRI MRITomato MRR=MRI

Sources: Extracted from Hills, B.P. and Clark, C.J., Annu. Rep. NMR Spectrosc., 50, 75–120, 2003; andcomplemented by Butz Jahreszahl fehlt; Létal, J., Jirák, D., Šuderlová, L., and Hájek, M.,

Lebensmittel-Wiss. Technol., 36, 719, 2003; Hernández-Sánchez, N., Barreiro, P., Ruiz-Altisent,M., Ruiz-Cabello, J., and Fernández-Valle, M.E., Appl. Magn. Resonance, 26, 431, 2004;Thybo, A.K., Szczypinski, P.M., Karlsson, A.H., Donstrup, S., Stodkilde-Jorgensen, H.S., and

Andersen, H.J., J. Food Eng., 61, 91, 2004; Chayaprasert, W. and Stroshire, R., Postharvest Biol.Technol., 36, 291, 2005; Gambhir, P.N., Choi, Y.J., Slaughter, D.C., Thompson, J.F., andMcCarthy, M.J., J. Sci. Food Agric., 85, 2482, 2005; Hernández, N., Barreiro, P., Ruiz-Altisent,

M., Ruiz-Cabello, J., and Fernández-Valle, M.E., Concepts Magn. Resonance Part B: Magn.

Resonance Eng., 26B, 81, 2005; Marigheto, N., Duarte, S., and Hills, B.P., Appl. Magn.

Resonance, 29, 687, 2005; Raffo, A., Gianferri, R., Barbieri, R., and Brosio, E., Food Chem.,

89, 149, 2005; Hernández-Sánchez, N., Barreiro, P., and Ruiz-Cabello, J., Biosystems Eng., 95,529, 2006; Brescia, M.A., Pugliese, T., Hardy, E., and Sacco, A., Food Chem., 105, 400, 2007;Goñi, O., Muñoz, M., Ruiz-Cabello, J., Escribano, M.I., and Merodio, C., Postharvest Biol.Technol., 45, 147, 2007; Hernández-Sánchez, N., Hills, B.P., Barreiro, P., and Marigheto, N.,

Postharvest Biol. Technol., 44, 260, 2007; Tu, S.S., Young, J.C., McCarthy, M.J., and McCarthy,K.L., Postharvest Biol. Technol., 44, 157, 2007.

Note: MRR, nuclear magnetic resonance relaxometry; MRS, nuclear magnetic resonance spectroscopy;

MRI, nuclear magnetic resonance imaging; PD, proton density.


Most of the studies referred to in Tab le 5. 3.1 focused on frui ts, such as apple todetect internal brow ning, mealines s, bruis ing, and watercor e by means of NM Requipment operat ing at magne tic field strengths from 0.13 to 4.7 T, and makinguse of relaxo metry and ima ging techni ques. The se techni ques have also beenimplem ented in a numbe r of studi es on inte rnal browning in pears. Matu rity e valu-ation ha s been the targe t for a numbe r of studies in cheri moya, duria n, kiwifr uit,mandarin, mango , pineap ple, and tomato. In these cases, as for previ ous applestudies, diff erences in relaxatio n times and in signa l intensities within MR imagesallowed the charact erization of the samp les.

NMR spectroscop y has been used to quanti fy chemi cal c ompounds such assoluble solids in a variety of speci es such as kiwifruit , melon, waterm elon, orange,grape, and cherry; and oil in avocado. Detection of inte rnal structure s has alsoachieve d encoura ging results by means of one-dimens ional (1D ) images for pits incherries or olives, and by means of two-dimens ional (2D) MR ima ges for seeddetection in citrus.

Most of aforement ioned wor ks have been undertak en with comm ercial NM Requipment designed for medical purpos es, whi ch are not concei ved to deal with thepractical const rains that apply for the food industry. Su ch equipm ents operat e athigh-magne tic field strength of more than 2 T with high-perfor mance hardware,which invol ves dif ficulties of setting in an industri al environ ment and large inves t-ments. In most of the cases, samp les remained stationar y durin g data acquis ition,which, on the other h and, was perfor med without time-con sump tion rest riction,being only of use for off-line exami nation. Practical on-line moni toring requiressample inspe ction under mot ion condition s and shortening of data acquis ition time.Neverthel ess, these studies compr ise a reveal ing approac h to a mul tisensi ng tool forfruit a nd vegeta ble inspectio n.

Only a few studies focus o n the effect of sample motion, reduct ion of the dataacquisition time, or use of low-m agnetic field stre ngth. How ever, the results obtainedhighlight the great potential of NMR techniques for internal quality monitoring andencourage further works for developing feasible on-line NMR systems.

The following two sections provide the basics of the NMR phenomenon and theNMR signal acquisition as the basis of the characterization of quality attributes,unfamiliar readers may directly skip to Section 5.3.4 where further explan ation isprovided for the quantification of success for most relevant applications, togetherwith some transferability remarks with regard to industrial needs.

5.3.2 BASICS OF NMR RELAXOMETRY AND NMR SPECTROSCOPY

5.3.2.1 Magnetic Moment of Nucleus and Its Excitation

NMR is based on the phenomenon that nuclei consisting of unpaired nuclearparticles, that is, odd number of protons and neutrons possess an intrinsic propertycalled spin and therefore have angular momentum. The angular momentum and thenuclear charge confer a magnetic moment on the nucleus, which can interact with anexternal magnetic field. Such nuclei include 1H (proton), 2H, 13C, 15N, 19F, 23Na,31P, etc., among which 1H represents the highest quantity and biological abundance


mainly as protons in water molecules, and the highest NMR signal. Proton NMR is avaluable tool for fruit quality evaluation as water influences many quality-relatedcharacteristics and metabolic processes, and therefore, we will focus on suchnucleus.

The interactions of the atoms with external magnetic fields can be described byquantum mechanics, although the overall behavior is characterized in classicalphysics. A combination of both descriptions is normally used to illustrate thephenomena, and the latter will be what we adopt in the next paragraphs.

In the absence of an external magnetic field, the magnetic moments of thenuclei are randomly oriented having the same energy and zero net magnetization.In the presence of a static magnetic field B0, the magnetic moments will line up withthe magnetic field at different angles determined by the spin number (Figure 5.3.1).The possible orientations correspond to different energy levels being the energydifferential directly related to the static magnetic field strength.

For 1H, two orientations are possible: parallel and antiparallel. Parallel protonorientation corresponds to a slightly lower energy level in comparison to antiparallelproton orientation (Figure 5.3.1). At thermal equilibrium, a slight excess of protonsresides in the parallel state giving rise to a net macroscopic magnetization (M), whichproduces the NMR signal. Curie’s law (Equation 5.3.1) establishes the equilibriumlongitudinal magnetization as function of the static magnetic field, B0, through thestatic nuclear magnetic susceptibility, x0, which is characteristic of each nucleus andinversely dependent on the temperature.

M ¼ x0 B0 (5:3:1)

The macroscopic magnetization increases as the magnetic moments align with theexternal magnetic field until the thermal equilibrium is achieved. The characteristictime constant for the equilibrium establishment is T1, the so-called longitudinalrelaxation time. The net macroscopic magnetization, M, is the sum of the contribu-tions of all the magnetic moments of the individual protons. The component parallel

y

yz

z

B0

M

x

x

FIGURE 5.3.1 On the left, magnet axis established by convention. On the right, netmagnetization (M) along the direction of the main magnetic field (B0) as a result of the paralleland antiparallel nuclear magnetic moments orientation.



to the magne tic field direc tion is called longitudinal magne tiza tion; from nowonward we will refer to it as Mz .

Because of the angula r mom entum each magne tic moment undergo es a torqu eand preces ses a round the stat ic magne tic field axis describing a circu lar orbit. Theangular freque ncy o f nu clear preces sion ( v0) is known as the Larmor freque ncy(Equation 5.3.2) and is direc tly propor tion al to the external magne tic field strengthvia the gyrom agneti c ratio ( gi ), which is dependen t on the nucleu s. The gyromag-netic ratio descri bes the relat ionship between the nucleu s-spe cifi c magnetic momen-tum ( ~m) and ang ular momentu m (~S), ~m ¼ g~S (Table 5.3.2).

v0 ¼ g i B0 ( 5: 3:2)

The behavi or of the magne tic momentu m in the presen ce of a tim e-independentmagnetic field ~B ¼ B0~z is well know n and can be described by the Schrödi ngerequation. The solution for this equati on just ifi es the use of classical preces sion forthe proton spin motion. Moreover, the discretene ss of the proton ’ s intr insic ang ularmomentu m leads to the discretene ss of the energy levels o f its interacti on with amagnetic field, and thus to the paral lel and anti parallel states for a proto n (or any spin1/2 nuclei ). Thi s is an e xample of the general Zeeman effect, wher e nuc lear magne ticmoment in the presence of an external magne tic field leads to splitt ing in nuc learenergy level s.

The phenom enon of resona nce occurs when an electrom agnetic pulse is appliedwith energy enough to induce transitions betw een these states. The freque ncy atwhich the nucleu s will absorb such energy is preci sely the Larmor frequency. Inproton NMR, the necess ary energy evolve s in the radio freque ncy (R F) range.Excitation RF pulse is appli ed wi th an oscillat or coil by produci ng an alte rnatingfield B1 perpendicular to B0 with much smaller magnitude compared to it. In classicalphysics, the behavior of the nuclei is described such as the longitudinal magnetiza-tion precesses around the B1 axis at the angular frequency v1¼ g � B1. Here, the netmagnetization moves from the z-axis toward the transversal plane, wheresignal measurem ent is carri ed out (Figure 5.3.2). Depending on the product of the

TABLE 5.3.2Gyromagnetic Ratio g of Some Nucleiof Interest in Megahertz per Tesla

Nucleus gi [MHz=T]

1H 42.582H 6.5413C 10.7115N 3.0819F 40.0823Na 11.2731P 17.25


z

Detector coil

RF coil

M

Mx

My

I

It

yx

FIGURE 5.3.2 On the left, depiction of the spiral-like progress with time of the netmagnetization vector (M) in x–y plane where RF pulse is applied by the RF coil along the xdirection, and signal is detected by the detector coil along the y direction. On the right,evolution along the time of the x and y components of M after an RF pulse, note that Mx iszero and My is maximum after excitation.

amplitude and length of the applied RF pulse the so-called flip angle varies. A 908RF pulse refers to a right flip angle so that the longitudinal net magnetization (M) tiltscompletely on x–y plane. According to quantum mechanics concept, when the RFpulse reaches sufficient strength or duration, an equal distribution of nuclei at the twoenergy states is achieved so that the longitudinal magnetization disappears. Inaddition to the transitions between the energy levels, immediately after the RFpulse, the precession of the nuclei is synchronized and, as a result, a magnetizationperpendicular to the axis of the external magnetic field appears.

The magnetization on x–y plane is called transverse magnetization (Mxy) and thisis the component susceptible of being detected in NMR analyses. The magnitude ofthe NMR signal is proportional to the number of proton nuclei in the tissue andprovides a description of the latter in terms of density and composition as it reflectsthe distribution of air gaps, water, and metabolites.

5.3.2.2 Relaxation of Nucleus after Excitation

When the RF pulse is turned off, the magnetic moments return to the parallel lowerenergy level with simultaneous emission of energy at the Larmor frequency. Thus,the net magnetization swings back toward the positive z-axis until the equilibriumlongitudinal magnetization is recovered. This process is called longitudinal or spin–lattice relaxation, referring to the exchange of energy in the form of thermal motionsbetween the spins and their surroundings. To transfer the excess of energy, there is aneed of receivers that allow receptivity (tuning) to the precession frequencies of theexcited nucleus.

The rate of return to the initial equilibrium state is exponential, characterized bythe constant T1, longitudinal relaxation time (Equation 5.3.3). At T1, the longitudinalmagnetization has recovered 63% of its initial value M0. The magnitude of T1directly depends on the static magnetic field B0.

Mz(t) ¼ M0(1� e�t=T1) (5:3:3)


Longitudinal relaxation time T1 is different for different tissues and is a valuablesource of contrast in quality-evaluation measurements. Differences arise as a resultof the different environments at the microscopic and molecular levels, giving rise to achanged magnetization characteristics, which may be described by a multiple expo-nential decay function. When molecules in the actual environment have a movementrate or tumbling close to the Larmor frequency, the energy exchange is favored. Thelongitudinal relaxation becomes more efficient, that is, T1 shortens. As differences inthe thermal motion increase, efficient exchange is not allowed, which involves asubsequent T1 lengthening. Such differences appear in relation to the molecularmobility and size. Small molecules, such as inorganic salts and water, tumble ormove at much higher rates than the Larmor frequency being free to collide andso facilitating energy transfer. In contrast, larger and more rigid molecules, such aslipid, fat, protein, and complex sugars, move slower preventing collisions andso energy dissipation. The distribution of the molecular motions depends on tem-perature and viscosity. T1 is expected to increase at lower temperature and for moreviscous media.

As the longitudinal component Mz recovers its original value, the transversemagnetization Mxy decays as a result of the loss of phase coherence of the excitedprotons. The cause is normally associated with magnetic field inhomogeneities. Overtime, the individual transverse magnetization vectors begin to cancel out and the netmagnetization vector gradually reduces. The rate of the decay is also exponentialwith the time constant T2, transverse relaxation time (Equation 5.3.4).

Mxy(t) ¼ M0 � e�t=T2 (5:3:4)

The origin of the loss of phase coherence is caused by two factors. On one hand,there is an external origin related to the imperfections in the magnetic field homo-geneity of the NMR spectrometer. The second one refers to mechanisms of T2relaxation inherent to the sample.

It is worthy to note that since the external magnetic field is rarely completelyhomogeneous, an effective relaxation time shorter than T2 governs the decaying, thisis the T2*. The difference resides in that the dephasing induced by the external fieldinhomogeneities can be reversed, whereas T2 effect is not reversible.

As for the inherent causes of loss of phase coherence, protons can be regarded asmagnetic dipoles that cause intrinsic magnetic fields within the sample with inten-sities proportional to their magnetic susceptibility. Magnetic susceptibility variationsand interactions among the nuclei themselves create local inhomogeneous magneticfields. At the molecular level, internal and tumbling motions of the molecules andchemical exchange between proton pools characterized by different precessionfrequencies are sources of T2 relaxation. Diffusion and tissue microstructure presenttheir effects at a microscopic level.

In a viscous or solid medium, molecules are relatively fixed, which involves thepresence of relatively fixed local magnetic fields that cause local inhomogeneities inthe protons surroundings. This environment could lead to differences in precessionfrequencies that increase the rate of dephasing. In liquids, the local magnetic fieldsfrom neighboring molecules fluctuate rapidly being averaged to a small value so that


dephasing is less favored and, consequently, T2 is longer. Proton exchange occursbetween water and exchangeable protons on cell metabolites, such as sugars, cellwall biopolymers, or starch granules (Hills and Clark 2003). A slow rate of protonexchange between chemical sites causes the vectors to loose phase coherencebecause of local magnetic field differences. Viscosity facilitates proton exchangeas nuclei are held in close proximity for an extended period.

The evolution of the metabolites concentration during fruit and vegetable ripen-ing exerts an important effect on the relaxation times. For unripe fruit, starchgranules are major relaxation sinks for water protons. As fruit ripen, starch granulesbreak down into sugars, decreasing the size and the number of granules. However, T2decreases since more hydroxyl groups are available to exchange with the waterprotons (Keener et al. 1997).

At a microscopic level, cell characteristics and tissue microstructure present amajor influence on the transverse relaxation process, which is associated with theresultant effect of the water diffusion between subcellular organelles and intercellulargaps. Consequently, T2 is a powerful source of information on cellular structureintegrity. Water is normally compartmentalized into vacuole, cytoplasm, and extra-cellular space with increasing mobility restriction, respectively, that leads to decreas-ing T2 values (Hills and Remigereau 1997). Depending on the cell morphology andsize, and the membranes permeability, the diffusion of water between compartmentswill bring different levels of magnetization averaging and hence, loss of the contrastinformation. Similarly, loss of membranes integrity results in changes in watercompartmentation and in overall T2 values, which could be detected by analyzingthe distribution of the transverse relaxation times and the influence on macroscopiccontrast. When water diffuses through regions with different magnetic susceptibi-lities, the local field gradients that appear as consequence of such discontinuitiesproduce magnetization dephasing and consequently T2 shortens. This situation takesplace especially at air–liquid interfaces and it is enhanced at high-static magneticfield strengths, since local field gradients are intensified. Table 5.3.3 summarizes the

TABLE 5.3.3Summary of the Main Parameters Affecting the Time Constantsthat Characterize the Longitudinal and the Transverse RelaxationProcesses of the Magnetic Moments (T1 and T2, Respectively)

Longitudinal Relaxation Time (T1) Transverse Relaxation Time (T2)

Molecular motion Molecular motionMolecular size Chemical exchangeMolecular complexity Cell morphology and sizeViscosity Cell compartmentation

Membrane permeabilityTissue microstructureDiffusion


parameter s affecting relaxatio n times. For further readi ng on NM R relax ation in foodproducts, the boo k by Hills (1998) is recom mende d.

5.3.2.3 Signal Dete ction dur ing Relaxat ion

Followi ng the RF pulse, the precessing trans verse magne tization Mxy generat es asmall elect romotiv e force that , ac cording to Fa raday ’s law of induction , induce s acurrent along the recei ver coils. Su ch coils are situated in the trans verse plane (F igure5.3.2) and separa te out the signals along the x and y axes. Acco rding to thei r shape,there are volume and surface coils. The same coil can be used to transmit and toreceive. The receiver coil observes an oscillating wave signal with a phase describedby (ivt), where v is the nucleus resonance frequency (Figure 5.3.2). The signalamplitude decreases to zero in exponential fashion with time as exp(�t=T2). Thedetected time-domain signal is called free induction decay (FID) because of itsdamping nature. The signal is recorded as an oscillating voltage, which is amplifiedand digitized by an analogue-to-digital converter, and then stored as an array ofcomplex data, which contains the information on the magnitude and the phase of thereading.

5.3.2.4 NMR Relaxometry and NMR Spectroscopy

As aforementioned, the magnetization of hydrogen nuclei in different physical orchemical environments decays at different rates. Moreover, different chemical envir-onments of otherwise equivalent hydrogens lead to slight differences in their pre-cession frequencies. MRR identifies nuclei populations distinguishable on the FIDbecause of the different decay time constants. Values can be computed for local spotsas well as it is possible to spatially resolve the assignation of times. For the latter, theoutcome is the so-called relaxation maps. As for MRS, the precession frequencyencodes the chemical groups that give rise to the NMR signal. The outcome is anNMR spectrum where intensity is plotted versus frequency.

5.3.2.4.1 NMR RelaxometryThe determination of the longitudinal relaxation time T1 is usually performed withthe inversion-recovery sequence. The sequence consists of a 1808 RF pulse thatinverse the longitudinal magnetizationMz. During an inversion time interval (TI),Mz

is allowed to recover. Then, it is tilted toward the transverse plane by a 908 RF pulseand the magnitude of M�z is measured, which is proportional to the PD and T1. Thisprocess is repeated for a number of different TIs. After several measurements, theconstant time T1 can be determined from Equation 5.3.5.

MT1 ¼ Mz(1� 2e�TI=T1) (5:3:5)

T2 determination is implemented by means of the Carr–Purcell–Meiboom–Gill(CPMG) sequence. First, a 908 RF pulse is applied to flip the longitudinal magnet-ization toward the transverse plane. After a time interval (t), a train of 1808 RFpulses is applied with a pulse spacing of 2t. The 1808 RF pulse refocuses the


dephased transverse magnetizations as the slower precessing vectors are now in frontof the faster precessing vectors, so that the latter overtake the slower ones. This factcauses what is called spin echo (SE) with the vectors reachieving coherence, losing itafterward. Maximum signal is achieved at time interval t after each refocusing pulseapplication. The peak intensity is proportional to the PD and T2. The envelope of thepeak echo intensities decays along the echo train that allows calculation of T2(Equation 5.3.6).

MnTE ¼ Mze�tnTE=T2 (5:3:6)

The effects of the inhomogeneities in the magnetic field on the loss of phasecoherence, that is, the effective T2 (T2*) decay, are cancelled out.

When more than one single relaxing component (as those corresponding tovacuoles, cytoplasm, and intercellular spaces) are expected, deconvolution of theecho decay envelope is done either by assuming higher order exponential decay orby presenting the data as a continuous distribution of relaxation times and decon-volution with an inverse Laplace transform (Hills et al. 2004). See example onFigure 5.3.3. Thus, MRR also provides microstructure information on the sample.

16,000

14,000

12,000

10,000

8,000

6,000

Sig

nal [

a.u.

]S

igna

l [a.

u.]

4,000

2,000

2,500

2,000

1,500

1,000

500

00.01 0.1 1 10

T2 [�105 ms]

Time [�106 ms]

100

00 2 4 6 8 10

FIGURE 5.3.3 Example of a relaxation curve measured on apple tissue (top) and result of T2continuous distribution fitting (bottom) where two peaks are clearly resolved corresponding todifferent subcellular compartments, that is, water in vacuole and in cytoplasm, respectively.(Reprinted from Barreiro, P., Moya, A., Correa, E., et al., Appl. Magn. Resonance, 22, 387,2002. With permission.)


Relaxom etry p rovides basic infor mation regard ing tissue charact eristics that canreadily be applied for enhanci ng MRI contrast. Besides it may be directly employedas a mean for whole frui t quality assessmen t and for tissue microst ructure analys is, asit will be discussed in Se ction 5.3.4.

5.3.2.4.2 NMR SpectroscopyThe differences in the precession frequencies in different chemical environmentsoriginate in the electron orbitals that appear around the nuclei. The circulation ofelectron probability in the orbitals causes a small magnetic field at the nucleus thatopposes the externally applied one, this is the nuclear shielding. Thus, the effectivemagnetic field affecting the nucleus is generally less than the applied field by afraction s (shielding constant) as indicated in Equation 5.3.7.

B ¼ B0(1� s) (5:3:7)

The electron density around each nucleus in a molecule varies according to the typesof nuclei and bonds in the molecule. Therefore, protons with different chemicalenvironments will resonate at slightly different frequencies from that defined by theapplied external field B0. This is called the chemical shift phenomenon.

The chemical shift of a nucleus (d) is the difference between the resonancefrequency of the nucleus (n

_) and a reference resonance frequency (nREF) divided by

nREF (Equation 5.3.8). Since the resonance frequency depends on the magnetic fieldstrength, the relative scale removes the field dependence so that comparison betweendifferent NMR equipments becomes feasible. Chemical shift value is very small andit is generally reported in parts per million [ppm] (Table 5.3.4).

TABLE 5.3.4Characteristic Proton Chemical Shifts for FunctionalGroups of Interest

Functional Groups Structure Chemical Shift [ppm]

Aromatic Ar��H 6.0–8.5Benzylic Ar��C��H 2.2–3.0

Alcohols H��C��OH 3.4–4.0Ethers H��C��OR 3.3–4.0Esters RCOO��C��H 3.7–4.1

Esters H��C��COOR 2.0–2.2Acids H��C��COOH 2.0–2.6Carbonyl compounds H��C��C¼¼O 2.0–2.7

Aldehydic R��(H��)C¼¼O 9.0–10.0Hydroxylic R��C��OH 1.0–5.5Phenolic Ar��OH 4.0–12.0

Enolic C¼¼C��OH 15.0–17.0Carboxylic RCOOH 10.5–12.0Amino RNH2 1.0–5.0


Avocado

Plum

Water peak

Oil peak Sugarpeak

30,000

20,000

10,000

0−4 −2 0

Mag

nitu

de [a

.u.]

Chemical shift [ppm]

2 4

FIGURE 5.3.4 NMR spectra of an avocado and a plum showing the water, sugar, and oilpeaks at the chemical shift of the corresponding protons. The horizontal axis represents thechemical shift expressed in parts per million as a relative difference in frequency from theresonance frequency of water. (Courtesy of P. Chen, unpublished data.)

d ¼ (n � nREF)106=nREF (5:3:8)

The scale of the chemical shift establishes a relation between position and atomor group of atoms, which allows the identification of different compounds in theanalyzed sample on the basis of frequency information (Figure 5.3.4).

Such frequency information is extracted from the time-domain signal, by apply-ing the Fourier transformation (FT) under its abbreviated form (fast Fourier trans-form, FFT) to the FID. The measured time-domain signal is a superposition ofindividual FIDs and contains all of the frequencies coming from the excited sample,which are measured simultaneously. The FT is a mathematical tool that separates thecontribution of each nucleus by its resonance frequency and identifies the corre-sponding intensities of the components of the FID. As we said above, the frequency-domain spectrum is a plot of intensity versus frequency (Figure 5.3.4). The areas ofthe resulting spectral peaks are proportional to the nuclei concentration and numberof chemically equivalent nuclei of each molecule. The peak widths are inverselyrelated to their transverse relaxation times and play a key role in peaks resolution.For most fruits, the resolved types of hydrogen nuclei are associated with water, oil,and carbohydrates.

5.3.3 MRI FUNDAMENTALS

This section describes the obtention of images by means of spatial encoding usingmagnetic field gradient, a smart idea that has became the basis of several Nobelprizes, and which has led to a full discipline. It requires some skill to approach the


underlying physi cal and mat hematical concept s for a deep unders tanding of theissue. Practical approac h is also widespr ead by means of simp le compa risons ofimages correspondi ng to characte ristical samples. The form er aspect s are describedin this section, while the latter are detailed in Secti on 5.3 .4.

5.3.3.1 Image Acquisition and Reconstruction

For imaging, the NMR signal must carry the spatial information related to thelocation within the magnet. As illustrated in Equations 5.3.9 through 5.3.11, it isdone by the application of a linear magnetic field gradient (G) so that the magneticfield becomes dependent on the position (s) and so does the frequency at which thespins precess (vs).

Bs ¼ B0 þ G � s (5:3:9)

vs ¼ g � Bs (5:3:10)

vs ¼ g � (B0 þ G � s) ¼ v0 þ g � G � s (5:3:11)

When the frame of reference rotates at the angular frequency of the spectrometer, thatis, v0¼ g B0, the frequency of precession at position s is given by g<G< s. Theresultant phase Fs is a function of the length of time during which the gradient is on.

In the conventional MRI sequences, three linear magnetic field gradients areimposed along the x, y, and z directions, respectively, to spatially encode the signal.When selecting a slice perpendicular to the main magnetic field direction, the firststep consists of turning on a field gradient (Gs) along the B0 direction, which is the zdirection by convention. At the same time, a slice-selective RF pulse is applied toexcite the spins precessing in a range of frequencies determined by the bandwidth ofthe RF pulse and the gradient strength. The selected plane refers to the field of view(FOV) whose characteristic features are the in-plane dimensions in square or rect-angular geometry and the thickness.

The selection of the plane on which the FOV lays is carried out to assure that theregion of interest is included in the image. Three main orientations are generallyselected for the FOV. According to medical literature, slices that lie on the x–y planeare called axial or transversal, those lying on the y–z plane are called sagittal, andthose on the x–z plane are called coronal. No further implications are involved unlessthe sample is in motion during image acquisition. Such conditions will be discussedlater in this section.

Immediately after the first slice-selective RF pulse, a gradient along the ydirection (phase encoding gradient, GF) is switched on making rows within theFOV (and perpendicular to the gradient direction) to precess at different frequencies.After a time interval t the gradient is switched off. The spins return to precess at thesame frequency but the phase differences between rows persist. Next, a gradientalong the x direction (frequency encoding or readout gradient, Gf) is applied duringsignal acquisition. This gradient makes each FOV column to precess at differentfrequencies, while the phase differences achieved in the phase-encoding direction arepreserved. Thus, an echo is generated. The number of sampled points within the echoequals the number of encoded columns.


The volume elements identified by pairs of row and column numbers are calledvoxels. Their phase will be imposed by their location within the FOV (x, y). Thereceived signal over time is the sum throughout the FOV of the transverse magnet-izations at each voxel with their corresponding encoded phases (Equation 5.3.12).

S(t) ¼ð ð

m(x, y)e�igGFyte�igGfxt dx dy (5:3:12)

wherem(x, y) is the transverse magnetization of the voxel under spatial coordinates xand y

gGFyt and gGfxt refer to the acquired phase owing to the phase and frequencyencoding, respectively (Figure 5.3.5)

This timing schedule is repeatedover timeat intervals termed repetition time (TR).Aftereach RF pulse, the intensity of the phase encoding gradient is made to vary obtainingfinally as much echoes as number of encoded rows. The image information is notencoded directly, the digitized values of each echo are sequentially stored in rowsresulting in a 2D array, the so-called k-space (the k is by analogy wavenumber) orreciprocal domain,whichcanbe regardedasamapof spatial frequencies (kxand ky)withthe correspondence indicated in Equations 5.3.13 through 5.3.15. The k-space presentsconjugate symmetry for positive and negative frequencies (in mathematical operationswith complex notations, signed frequencies stand for rate and direction of rotation).

S(t) ¼ S kx(t), ky(t) �

(5:3:13)

kx(t) ¼ðt0

gGf (t) dt (5:3:14)

FOV

2D-FFT

x

y

Frequency-encodingmagnetic field gradient

Pha

se-e

ncod

ing

mag

netic

fiel

d gr

adie

nt

FIGURE 5.3.5 Schematic of signal spatial codification using linear magnetic field gradientsalong the x and y directions of the FOV and example of a reconstructed MR image afterapplication of a 2D-FFT.


ky ky

kx kx

kf

FIGURE 5.3.6 k-space sampling by Cartesian coordinates (left) and polar coordinates(right). For the latter data, interpolation is required before image reconstruction to obtain aCartesian grid.

ky ( t ) ¼ðt0

g GF ( t ) dt ( 5: 3:15 )

For convent ional MR I sequen ces, the magne tic field gradients are appli ed such thatk-space is fi lled in a rectangular Cartesian way. In other sequenc es, the reciprocaldomain data a re coll ected using spiral or radia l trajector ies (Figur e 5.3.6). Advan-tages and disadvantag es are discu ssed late r in the text . The tota l time required tocomplete data acquis ition ranges from hours to tens of millisecond s dependi ng on thetype of acquisition param eters (num ber of phase cycles and numbe r of cycles perTR) and hardw are speci fications of the equipm ent. To enhance the applicabilit y ofMRI techni que for monitor ing purpos es, a compr omise betw een total acquis itiontime, c ontrast, an d spati al resoluti on has to be reache d.

A 2D MR image is recons tructed by applyi ng 2D fast Fourier transform ation(2D-FFT ) to the k-space . The resultin g ima ges are mat rices, usual ly 2D (one ormultiple) and 3D, consisti ng of an array of pixel s whose intensity depends onacquisition parameter s and diff erent proper ties, such as the PD contained in thesample, diffusion , relaxatio n proper ties, etc. (Figure 5.3.5).

5.3.3.2 Effect of Movem en t on Image Quality

Internal quality evalua tion under on-line condition s implie s that the samp les areconveyed throug h the magne t along the direc tion of the main magne tic field(z direction ). When the v olume to be excited by the RF pulse s, the FOV, lies onthe axial location (x– y plane in our schem e of magne t), that is, perpend icular to themotion axis, signa l arising from adjace nt tis sue slices passi ng throu gh it is regis teredduring the acquis ition time. Miscellan eous signa l produce s blurring artifact in theimage that c annot be correct ed. Su ch ima ges only will be useful as long as the signa lsuperimpo sition does not conc eal the stru ctures of inte rest within the samp le underinspection owing to image blurring.

Coronal and sagittal locations of the FOV (x–z and y–z planes, respectively)assure the acquisition of the signal from the same slice of tissue for all the phase-encoding steps. In this case, the slice remains within the volume delimited by theFOV, although its position is changing during the acquisition since the sample


moves forward. Such motion produces a measurable phase shift in the raw data thatis susceptible of being corrected according to the Fourier shift theorem.

This theorem states that if a function f(x) has an FT, F(s), when f(x) is shifted Dx,the resulting FT, F0(s), will be e�i2pDxsF(s) (Bracewell 2000). That is to say thesignal from the moving sample is equivalent to that of the stationary sample F(s)multiplied by a phase factor e�i2pDxs, which depends on the displacement of theobject (Korin et al. 1989). Note that F(s) is a complex number [Rþ iI] and so is F0(s)[R0 þ iI0].

The phase shift (DF) at any point of the k-space is mainly a consequence of thesample motion between phase-encoding steps. The expression for its computation isanalogous for displacements along the phase-encoding direction and along thefrequency-encoding direction (Equation 5.3.16). Such displacements derive fromthe velocity of the moving object along the corresponding direction and the timeelapsed between consecutive phase encoding steps, n and t, respectively, in Equation5.3.16.

DF ¼ 2p � nt � s� N � 12

� �� 1N

(5:3:16)

For displacement in the phase-encoding direction, s is the number of the phaseencoding step while N is the total number of steps. For displacement in the fre-quency-encoding direction, s is the point within the echo and N is the total number ofecho points acquired. The equation establishes the reference of the phase shift in thecentral phase-encoding step and it increases gradually for the others. When the imageis reconstructed from the corrected k-space, the sample will be positioned at thelocation occupied during the acquisition of that reference step.

The corrected values (Rc and Ic) corresponding to the real (R0) and the imaginary(I0) components of the complex points comprising the k-space acquired duringsample motion are provided by implementing Equations 5.3.17 and 5.3.18.

Rc ¼ R0 � cos (DF)� I 0 � sin (DF) (5:3:17)

Ic ¼ R0 � sin (DF)þ I 0 � cos (DF) (5:3:18)

Under given on-line conditions, that is, at a given belt speed, given acquisitionsettings such as TR and given number of pixels in the image along the motiondirection, the induced phase shift is unique and straightforwardly computed. There-fore, correction algorithms can be generated for different experimental conditions sothat the correction procedure could be standardized. Thus, its use could becomeroutine to correct rectangular k-space of any MR image in the framework of on-lineapplications. The major requirement for the application of standard correctionalgorithms is the existence of a smooth and stable conveyor belt speed, sincevibration and belt jerks would cause image blur that are more difficult to correct.

Even though motion correction is theoretically feasible for any belt speed, thereare limiting FOV restrictions for increasing velocities. For coronal or sagittal images,the FOV has to be enlarged as to cover the sample displacement during signal


acquisition, which increases at higher speeds. Consequently, distance between sam-ples would have to be lengthened. For axial orientation, the distance covered by thesample while passing through the FOV thickness is shorter, which allows nearerplacement of the fruits, and hence, raise the number of fruits inspected per unit time.However, as aforementioned, the image blurring cannot be corrected for this type ofimages. Therefore, a compromise between increasing performance and decreasingimage quality should be achieved.

5.3.3.3 Sequence Parameters and Their Effects on MR Image Quality

The image quality with regard to the MRI techniques is defined by the contrast of theimages, the ability to spatially resolve detail, and the SNR (Woodward 1995). Thissection provides a summary on several parameters that affect such quality imagecharacteristics.

5.3.3.3.1 Effect on ContrastIn MRI, contrast is the difference in relative brightness between pixel units. Bright-ness is directly related to the signal intensity received from each voxel, which istranslated into gray or false color scale ranges (Woodward 1995). Several sequenceparameters affect the signal intensity differences between tissues, such as the TR, theecho time (TE), and the flip angle.

The TR is the TI between successive cycles of accumulations or phase steps.During this time, the longitudinal magnetization recovers and signal becomes avail-able to be flipped into the transverse plane. Therefore, long TR implies higher signal.However, tissue contrast is poor if all tissues have recovered a similar amount oflongitudinal magnetization. Therefore, to enhance tissue contrast through T1 differ-ences (T1-weightening), TR needs to be shortened. Thus, tissues with short T1 valuesrecover faster and contribute with higher signal in the next excitation, whilethose with longer T1 values will have varying recovering rates and so a variationin contrast.

The TE is the time elapsed between the excitation RF pulse and the maximumsignal of the formed echo. TE controls the amount of dephasing between transversemagnetizations, and hence, the loss of signal. Long TE results in lower signalalthough enhanced T2-weightening. Those tissues with long T2 time take a longertime to dephase so that they will appear brighter in the image than those with fasterloss of coherence. Moreover, the T2 effect predominates when TR is lengthened as T1contrast minimizes.

Flip angle is the angle rotated by the longitudinal magnetization toward thetransverse plane after RF excitation pulse. For short flip angles the T2 contrastpredominates. As flip angle increases, differences in longitudinal magnetizationrecovering between components intensify and those components with short T1 willappear increasingly brighter. This parameter is in close relation with TR and usuallytheir values are chosen accordingly.

There are other sequence-related parameters that affect tissue contrast that are notconsidered here, such as inversion time, diffusion parameters (diffusion gradienttime and strength), number of accumulations, presence of magnetization transfer or


magnetization saturation pulses, etc. Information on these can be found in otherreviews or book chapters (e.g., Bernstein et al. 2004).

5.3.3.3.2 Effect on Spatial ResolutionThe spatial resolution is an image characteristic related to the minimum size of anobject that can be identified and to the clear definition of edges and boundariesbetween regions (Woodward 1995). The intensity of the magnetic field gradients[T=m] used for in-plane spatial encoding of the signal within the FOV alongside theFOV thickness defines the number and dimensions of a series of volume elements intowhich the FOV is divided, which are called voxels. The smaller the voxel, the higherthe resolution appears, although a loss in the SNR is concomitant. For equal in-planevoxel dimensions, an increase in the slice thickness involves an increase in the signalas it arises from a higher excited volume. However, thicker slices may containdifferent tissues with different signal intensities so that their overlapped contributionslead to a signal misregistration during FT. This effect is called partial volume effect.

The FOV dimensions and the size of the raw data matrix also are involved in thespatial resolution. The former is selected to cover a particular tissue volume ofinterest. The latter is defined by the number of phase-encoding steps and the numberof sampled points required to yield a desired spatial resolution. For a certain spatialresolution, the sampling size and FOV need to be chosen accordingly with thecriteria given by the Nyquist rate to avoid image aliasing. The larger the matrixdimensions and the in-plane voxel dimensions, the larger the FOV. Changes in FOVwhile the matrix dimensions are the same affect the in-plane voxel size in such a waythat increasing FOV means increasing voxel size.

5.3.3.3.3 Effect on Signal-to-Noise RatioThe SNR is given by the relative contributions to a detected signal of the true signaland randomly superimposed signals (background noise). SNR is affected by manysequence parameters. Flip angles are related to the transversal magnetization after RFexcitation pulse. Thus, increasing flip angles produce increasing transversal magneti-zation. One common method to enhance the SNR is to average several acquisitionsor pulse sequence repetitions, as signal sums up and random contributions cancelout. The SNR increases with the square root of the number of acquisitions used foraveraging. The SNR can also be improved by sampling larger volumes, that is, byincreasing the FOV and slice thickness (with a corresponding loss of spatial reso-lution) or, within limits (e.g., relaxation properties, susceptibility artifacts, coils,etc.), by increasing the magnetic field strength of the NMR spectrometer. It is worthyto note that an improvement in SNR does not necessarily result in a net improvementof the image quality as SNR is inversely related to spatial resolution.

5.3.3.4 Fast and Ultrafast MRI Sequences

For food evaluation, speed is required under an economical interest as it increases theinspection performance in terms of inspected sample unit rate, whereas under atechnical point of view there is a need for minimizing the sample motion effects,which are present whenever an on-line inspection is carried out.


The basic imaging time equation for a conventional MRI technique is given byEquation 5.3.19.

Scan time ¼ TR� NPE� NA (5:3:19)

whereTR is the repetition time between successive excitation RF pulsesNPE stands for the number of phase encoding stepsNA refers the number of acquisitions to be averaged to enhance SNR

Therefore, the reduction of the total scan time may be approached with severalstrategies by managing the three parameters involved. The shortening of TR isfavored by the use of short flip angles, although it is limited by gradient strength,which is normally an expensive engineering challenge. A different approach is usingfractional or partial k-space filling to exclude redundant data. By acquiring fewerlines of k-space or, in other words, NPE is reduced, uncollected data need to berepopulated by performing a zero filling or by duplication of data taking advantageof the conjugate symmetry of the k-space. For zero filling, central lines of the k-spaceare acquired so that image contrast is not altered, whereas those corresponding to thehighest phase-encoding steps, which present weak intensity signal, are substituted bystrings of zeroes.

An important way of time reduction relies on the possibility of acquiringmultiple k-space lines per TR, which is also called segmented k-space filling andhas led to a family of sequences broadly recognized as echo train imaging. The scantime of these images is reduced by a factor that depends on the number of lines persegment obtained consecutively before the next excitation RF pulse. In addition,optimized k-space sampling techniques can contribute to speed acquisition times.Decrease in scan time may also be achieved by reducing the number of acquisitionsused for signal averaging (NA). In fact, under on-line monitoring conditions NAis restricted to 1. In addition, more than one sample could be imaged within thesame FOV.

The counterpart of the actions devoted to time shortening usually is associatedwith a decrease of the SNR and the spatial resolution or to an increase of suscepti-bility effects on signal decay.

5.3.3.4.1 One-Dimensional Image ProfilesOne-dimensional image profiles are obtained by sampling with a single linear fieldreadout gradient applied along the direction of interest. 1D FT provides a profilewhere each point corresponds to the signal average of the entire volume located atthe position encoded by the applied gradient. Tissue differences are reflected in theprofile, which may be useful for detecting the presence, or the absence, of internalstructures such as pits in fruits or foreign bodies in fluids. This acquisition is fasterthan 2D or 3D imaging because it does not require repetitive sampling with intensityvarying phase-encoding gradients. Therefore, motion-induced artifacts may beneglected, which enhances its applicability to dynamic measurements. In addition,the SNR is higher owing to the registration of signal from larger volumes. The main


disadvantage refers to the partial volume effect, which is greatly enhanced due to theaverage tissue signal so that small structures may be concealed by the surroundingtissue.

5.3.3.4.2 Fast Low-Angle ShotThe fast low-angle shot (FLASH) sequence combines a low flip angle RF excitationpulse and a reversed gradient instead of a 1808 RF pulse, which refocuses the spins toform a gradient echo. As the excitation pulse angle is lower than 908, the residuallongitudinal magnetization permits shorter TR since it is not necessary to wait forits recovery by relaxation. The result of these arrangements is a significant scantime reduction up to 100 ms acquisition time. The k-space is filled linearly with onesingle k-space line being sampled after each RF pulse. As data are sampled on arectangular grid, the reconstruction complexity is reduced. Such disposition facili-tates the computation of the signal phase shift caused by sample motion andthe application of correction procedures. However, the SNR is low. In addition,the T2*-weighting involves sensitivity to magnetic field distortions and to suscepti-bility effects, although internal quality inspection could take advantage ofsuch susceptibility effects since tissues contrast may be enhanced, which wouldfacilitate identification.

5.3.3.4.3 Echo Planar ImagingThe echo planar imaging (EPI) is the best example of echo train imaging (see above).The exceptional shortening in acquisition time that characterizes this sequence (up totens of milliseconds) is achieved by the possibility of collecting part or even all thedata points comprising the k-space after a single RF excitation pulse (compromisingcontrast, resolution, and image distortion). In the classical EPI pulse sequence, theimage is acquired from one or several FIDs by applying a constant weak phase-encoding gradient and creating a series of echoes with strong, rapidly switchedfrequency-encoding gradient. As we have said, with modern gradient and RFhardware, EPI is capable of producing a 2D image in only a few tens of milliseconds(Bernstein et al. 2004). Such short times significantly overcome the problem ofimage degradation caused by the sample motion. However, the technical require-ments in software and hardware such as those regarding the gradient speed, strength,eddy currents screening, controllers, amplifiers, etc. are significant. The total dataacquisition must be performed within the T2* of the tissue, otherwise the signal isdestroyed before sufficient information is acquired (Vlaardingerbroek and den Boer1996). In these cases, image artifacts are considerable and the only alternative is toreduce the number of echoes in the echo train. Thus, the spatial resolution is limitedby the number of gradient echoes that can be acquired during the T2* signal decay.Artifacts in EPI sequences are as k-space line carries a different T2*-weighting sincethe k-space lines are acquired at different times, which causes image blurring alongthe phase-encoded direction (Bernstein et al. 2004).

5.3.3.4.4 Rapid Acquisition Relaxation EnhancedRapid acquisition relaxation enhanced (RARE) is a good example of SE train. It wasthe first multiple SE based one to fill more than one k-space trajectory in a single


excitati on (P arikh 1992). This fact reduces the acquis ition time, which dimini shes themotion- induce d artifact s. RAR E empl oys a train of 180 8 pulse s to refocu s decayi ngechoes. The k-space filling can be accom plished in one shot by a long train of RFpulses or in multip le shots c onsisting of shorter trains of RF pulses. As EPI , thecontrast is strongly depen dent on T2 relax ation because of the appli cation of the 180 8refocusing pulses provi ding an excellen t T2-we ighted tissue contrast.

RARE images are less sensi tive to main magne tic field inhom ogeneities andtissue magne tic suscep tibili ty varia tions than EPI sequenc es (B ernstein et al. 2004).The recons truction is straight forward be cause of the recti linear k-space.

Some d isadvantage s are related to the T2 signa l decay, which causes ima geblurring in the phase- encodi ng direc tion, and the presen ce of the 180 8 pulse s thatlimit the minim um spaci ng betw een succes sive echoes (Parikh 1992). In addition,RARE sequenc e presen ts high requireme nts on softwar e and hardware such as alsofound for EPI images since fast appli cation of the 180 8 refocu sing pulse s is required.

5.3.3.4 .5 S piral or Radi al k-Space Acquisit ionSpiral k-space samp ling is accom plished by the combinati on of two incre asing ,oscillating gradi ents. This modal ity compl etely elimin ates rapid gradi ent switchingrequired in sequenc e s such as EPI , which means much less deman ding requi reme ntson gradi ent hardware. Here no echo procedu re is requi red (De lpuech 1995) and thereis no relevant differences between the kx-axis and ky-axis (Rodríguez et al. 2004).Data in k-space are identified by means of polar coordinates, that is, the angle w withrespect to the kx-axis, and the distance k with respect to the origin of the (kx, ky) frame(Figure 5.3.6).

In spiral acquisitions, the trajectory of each readout step always starts at theorigin of the k-space, and ends at its edge. Because of the geometry of the trajectory,the central region, where contrast information is contained, is rapidly acquired andoversampled. Such characteristics confer to the spiral and radial sequences robust-ness with respect to motion-induced phase errors, susceptibility, and T2* decay(Vlaardingerbroek and den Boer 1996), and for spiral acquisition allow ultrafastimaging with total acquisition time of tens of milliseconds with fine contrast.

As for EPI, traversing the complete k-space within the time T2*, while acquiringsufficient sampling points with one single spiral, is very difficult. Therefore, inter-leaved techniques are applied by acquiring a number of similar spiral arms aftermultiple shots, which are rotated with respect to each other (Vlaardingerbroek andden Boer 1996).

Radial k-space sampling can be considered a particular case of the spiralmodality. Rodríguez et al. (2004) reported the development of a new imagingsequence called combined spiral and radial acquisition (COMSPIRA) that allowschoosing between spiral and radial center-out k-space trajectories in two and threedimensions (Pérez-Sánchez et al. 2006). The new design includes two parametersthat allow interleaved acquisition and sequence switching. The first is a measure ofthe angular difference between one readout step and the next, and the second onedefines the angular difference between the beginning and the end of a readout step ina number of complete 2p radians, which is called spirality. For radial trajectoriesspirality is zero.


Data acquir ed along a spir al or radia l traj ectory entail speci al algorithms forimage recons truction, such as data interpola tion to map the spir al da ta to a Cartesiangrid.

5.3.4 DETAILED VIEW OF A PPLICATIONS IN FRUITS

In this section, a numbe r of applicati ons of MRS, MRR, a nd MRI have bee n selectedfor a detailed view concern ing a varie ty of fruits (avocad os, cherri es, oliv es, ap ples,pears, and citrus), as icons for the industrial secto r.

Some of the appli cations are focuse d on the class ifi cation of frui ts in a limitednumber of catego ries (seedl ess or seed-c ontaining citrus, pitted and unpit ted cher-ries), while other s face the quanti fication of disorder degree (freezing inju ry, inte rnalbrowning or breakdo wn, mealines s), or even the chemical compo sition of fruits (oilcontent in avoca dos). Also statistical features such a s class ific ation rates and correl-ation coef ficient are provi ded for ea ch ca se as a mean for the quanti fi cation ofsuccess.

Most o f the applicatio ns that wi ll be p resented in detai l and all the techniques(MRR, MRS , and MRI) have been inves tigated under an on-line con figuration that iswithout doubt a main industri al deman d, and a challenge for the practical interest ofNMR in the field of horti cultural quali ty asses smen t.

5.3.4.1 Maturity in Avoc ados

Chen and coworkers (1996) designed and built a conveyi ng syst em to evaluate high-speed MRS technique in frui ts and vegeta bles. Single-pul se spect ra were obtainedfrom moving avocados to determin e the feasibil ity of the techni que for quantifyi ngthe dry weight of the fruit at different conveyor speeds. The misalignment of the fruitwith respect to the RF coil under both static and dynamic conditions, and the effectof the belt speed on the spectra were analysed.

The experiments were performed with a 2 T NMR spectrometer using a surfacecoil (Figur e 5.3.7 ). The magne tic field was shimme d solely with the first sample. Thestatic experiments were used to evaluate the effect of the fruit misalignment withrespect to the center of the surface coil. The correlation between the oil=waterresonance peak ratio and the percentage of dry weight of the avocados was 0.897for zero misalignment. A simulation with 400 repetitions showed that a misalign-ment within �4 mm from the surface coil would not significantly affect the peakratio and hence, the correlation. Under motion conditions, the RF pulse was activatedwhen the fruit was at position ranging from �10 toþ10 mm. Results showed that thestrongest signal was acquired when the sample was centered. The belt speed variedfrom 0 to 250 mm=s and the shape of the spectra as well as the line width showedvery little change over mentioned speed range. This result supports that the fieldremains sufficiently homogeneous throughout the distance range even though shim-ming was only performed with the first sample. The correlation between dry matterand oil=water peak ratio showed an average value of 0.975 and a SD of 0.004.According to these results, authors concluded that the method was rapid and accurateand had a great potential for on-line sensing of avocados maturity.


Steppermotor

Conveyorbelt

Conveyordriving

belt Woodensupport

Surfacecoil

Magnet

Positiondetector

FruitRoller

FIGURE 5.3.7 Schematic of an MR sensor with fruit conveying system for on-line fruitquality sensing. (Reprinted from Kim, S.M., Chen, P., McCarthy, M.J., and Zion, B., J. Agric.Eng. Res., 74, 293, 1999. With permission.)

Kim and coworkers (1999) extended the previous work on avocados by using thesame NMR equipment and conveyor belt and similar experimental procedure. Theconveyor speed was varied from 0 to 250 mm=s and spectra were acquired at fivedifferent positions relative to the coil center, computing three different peak ratios.This study also showed that the maximum intensity and width of the peaks changedwith relative location and belt speed. These authors explained that, for reducedcoupling between sample and coil, the excitation of the nuclei of the sampledecreases, leading to a decrease in the efficiency of signal recording and thus to areduction of signal intensity together with an increase in the line width of the spectra.This fact was overcome by computing the peaks ratio, which revealed to be muchless sensitive to belt speed and position. Increasing belt speed and sample displace-ment led to slight decreases in correlation coefficients between peaks ratio and dryweight, from 0.970 under the best conditions to 0.894 under the worst conditions.These results indicated, as in previous studies, the small effect of the motion of thesample on the acquired signal when single-shot sequences are used, as well as thegreat potential of the technique for this application.

Different studies have alerted the problems of using surface coils for signalacquisition on avocados, since these involve the excitation of the volume of tissueclosest to the coil. As indicated by Pathaveerat and coworkers (2001), when suchprocedure is implemented for on-line systems either the composition of the sampleshould be uniform or the spatial variation in composition need to be known. Toanswer these questions, chemical shift images, consisting of a spatially resolvedhigh-resolution NMR spectrum, were acquired with a Bruker Biospec 7 T spectrom-eter from avocados ranging from immature to very mature. It was found that thewater distribution is higher near the skin and seed and lower in the middle of theflesh, and that the oil distribution increases toward the seed. As for the oil=waterpeaks ratio, water was found large along the middle ring of flesh near the seed and


lower to the periphery. These results highlighted the importance of the knowledge onthe distribution of the compound of interest within the sample that allows establish-ing optimum procedure settings for maximum accuracy.

5.3.4.2 Pit in Cherries and Olives

Zion and coworkers (1994) focused on real-time detection of pits in processedcherries by 1D MRI profiles. In this case, a superconductor magnet with a magneticfield strength of 2 T was used, together with a homemade volume coil, for theacquisition of NMR projections. A permanent magnet operating at 0.26 T was alsoused to determine the effective transverse relaxation times (T2*) of pits and flesh incherries as prospecting study for enhancing the signal.

Based on the relative size of flesh and pits, a computer simulation of theprojections of cherries with and without pits was performed with the aim of opti-mizing the shape of the projections for various slice thickness and various sliceoffsets.

For real applications, three cherries at a time were placed inside the magneticfield roughly aligned along the sagittal plane (Figure 5.3.8). In the first set ofexperiments, cherries were placed randomly, which is the most likely dispositionowing to the difficulties in selecting orientation when conveying small-roundedfruits. A second set was obtained from pre-oriented cherries with the whole axislocated perpendicular to the excited plane. The magnetic field was shimmed onlyonce since, as explained by the authors, this operation is time consuming and couldnot be repeated under on-line specifications. An SE pulse sequence was used with nosignal averaging. A devoted Matlab routine was implemented as to identify the peaksand valleys in the projections and to calculate their ratios enabling to overcome thoseproblems arising from noise, shape distortions, and missing flesh parts. The SNRobtained with only one signal accumulation was enough to obtain reliable results.

0 20 40 Inte

nsity

[arb

. uni

ts]

Inte

nsity

[arb

. uni

ts]

60Position [pixel]

80 100 120 0 20 40 60Position [pixel]

Pitted cherries

Unpitted cherries

NMR profile NMR profile

NMR image NMR image

80 100 120

FIGURE 5.3.8 Examples of MR projections of a set of three cherries with pits (left) and a setof three pitted cherries (right) where pronounced valleys are observed for the latter samples.On the bottom, MR images of the corresponding cherries. (Courtesy of B. Zion, unpublisheddata.)


The classification routine yielded an 88% and 63% of correct classification for pit-containing and pitted cherries, respectively, corresponding to randomly placedsamples. Logically, classification performance was improved for pre-oriented cher-ries, 96.7% of correct classified cherries either with or without pits. The simulationindicated that excitation of a narrow slice is preferable because it is less affected byshifts of the cherry from the excited plane, which is likely to happen under on-lineconditions.

The short acquisition time of the projections ranged between 10 and 15 ms. Suchfast data processing and the possibility of accommodating wider number of cherriesin a longer volume coil reveal the high potential of this procedure to be transferredto industry application. However, the effect of the motion was not addressed in thisstudy.

Kim and coworkers (1999) obtained 1D MRI profiles using the conveyor systempresented in Chen et al. (1996) for studies on avocado, although making use of avolume coil instead of a surface coil. As before, only one acquisition was made foreach projection. For comparison, cherries were placed with their axis both perpen-dicular and parallel to motion direction. Projections were acquired from 0 up to 250mm=s belt speed. Sagittal imaging plane was chosen with the frequency-encodinggradient applied along the direction of the motion. Increasing motion again involveda decrease in signal intensity owing to relaxation effects. In contrast, there was littleor no distortion in the shape of projections. The best results to discriminate fruitbetween pitted and unpitted cherries were obtained when the pit was orientatedperpendicular to the motion direction. The classification rule was based on a simplethreshold yielding no errors for whole cherries under motion conditions, whereas71.7% error was found for pitted cherries placed with their axis parallel to themotion, and 1.7% for pitted cherries placed with their axis perpendicular to themotion. The classification performance was very similar to that of the static condi-tions (1%, 70%, and 2% for the same samples and placements). Despite theseencouraging results, the authors pointed out the need for further improvements inthe conveyor design and control, as well as in the sorting algorithms.

Zion and the working group applied the same procedure to the detection of pitsin olives (Zion et al. 1997). Olives were conveyed at belt speeds up to 250 mm=sachieving classification errors in segregating pitted and non-pitted olives lowerthat 5%.

The use of profiles for the detection of pits in processed cherries and olives isjustified by the large proportion of volume occupied by the pit within the piece,which involves significant pronounced valley appearing for pitted units in compari-son with unpitted units. In contrast, the detection of small structures within largersamples (such as seeds in citrus) requires 2D imaging to avoid partial volume effect.

5.3.4.3 Internal Browning in Apples

Internal browning is a physiological disorder characterized by the developmentof brown discoloration areas throughout the cortex and core, and the formation ofcavities without any external symptom. A high concentration of CO2 and low levelsof O2 within the fruit are factors to induce this disorder. Thus, CA storage intensifies


the occurrenc e of the internal brow ning, although it may also appear durin g thepreharvest period. Other factors such as varie ty, growing regio ns, and culturalactivities are also implic ated in the develo pment of this disorder.

Clark and Burmei ster (1999) perfor med an MRI study on the develo pment ofbrowning in ‘ Braeburn ’ apples using a 1.5 T clin ical instrum ent. Se ries of SET1-weight ed images wi th an in-plane resol ution o f 0.35 mm=pixel (4 mm thic k) wereobtained from fruits d uring 28 days of CA storage under high CO2 concent ration.Nine ima ges at diff erent locations wer e acquir ed in each series with a tota l imagingtime of 7.6 min.

In damag ed apples , discr ete areas of fl esh appeared as p ixels wi th higherintensity signal when compa red to the backgro und tissue. The ir signa l inte nsityincreased over time, while the patche s eith er enlarged or merg ed to form largeraggrega tes. When MR ima ges were compa red with photog raphs of disse cted app lesafter the 28 days of stor age, the highes t MRI signal areas coinci de wi th the browndamaged tissue. The increasing MRI signa l was attributed to a decreas ing mobility asa result of the conce ntration of metaboli tes such as acetaldeh yde and ethano lproduce d by respi ratory processes. How ever, authors also highlighted the dif fi cultyof ima ge contr ast inte rpretatio n as changes in signa l could also derive from changesin proto n den sities.

González and cowo rkers (2001) extend ed the study on inte rnal browning inapples by ident ifying the individua l contr ibution of the PD, longi tudinal relax ationtime ( T1), and transverse relax ation time (T2) to the image contr ast. Experim entalwork was carrie d out with a 0.6 T magne t. SE MR images of ‘ Fuji ’ apples wereobtained along with T1, T2, and PD maps wher e a quanti tative value of the co rre-sponding param eter is displ ayed for each voxel.

MR images wer e c ompared to photog raphs captur ed after cutt ing in half thesame apples. Brown-dis colored areas in the pictu res showed goo d correlati on withthe regio ns in the MR ima ges presen ting inte nsity signa l different from that of thenormal tis sue. Differe nt regio ns were identi fied on the basis of diff erent ranges ofintensities . Dar k brown and light brow n areas show high- and low-si gnal MRIintensity, respective ly, when compa red to sound tis sue. Characte ristic T1, T 2, andPD values were compu ted for each regio n by averagi ng the resul ts obtained fromthese areas in the correspond ing maps . In damag ed tissue, a reduct ion was observ edin both the T1 and the PD. As for T2, ligh t brown regions had short er T2, whereas thedark brown regio ns presen ted the highest relax ometric values . These values alongwith the experimental imaging parameters allowed calculating their relative contri-bution to the signal of each region. It was found that the transverse relaxation timecontributed more to the contrast than PD and T1.

To reduce the total acquisition time, resolution was diminished in the phase-encoding direction driving to a significant shortening in image acquisition time oftwo apples from 5.5 min to 20 s. Despite the increasing image blurring, theidenti fication of affected regio ns was sti ll feasible (Figure 5.3.9).

Jung and coworkers (1998) proposed a different approach for the detection ofinternal browning in apples. In this case, a global parameter, the transverse relaxationtime (T2), is computed to characterize the whole sample, instead of doing it regionallyon each tissue disorder. Apples were placed in a 0.13 T permanent magnet and a


(A) (B)

Dar

k br

own

Ligh

t bro

wn

Nor

mal

Noi

se

(C) (D)

(E) (F)

FIGURE 5.3.9 Example of SE MR images (A, C, and E with total acquisition times of 5.5min, 40 s, and 20 s, respectively) acquired from an apple affected by internal browning. Thecorresponding contour maps (B, D, and F) were used for the identification of normal, lightbrown, and dark brown tissues. (Reprinted from González, J.J., Valle, R.C., Bobroff, S.,Biasi, W., Mitcham, E.J., and MacCarthy, M.J., Postharvest Biol. Technol., 22, 179, 2001.With permission.)

CPMG pulse sequence was used to compute the overall T2 value. Results showed thatthe average T2 value of the apples with internal browning decreased as the severity ofthe disorder increased. T2–CPMG relaxation curves were fitted to a three-exponentialmodel to associate T2 values and proton relative populations with three differentcompartments, that is, the vacuole, the cytoplasm, and the extracellular space. Differ-ences in proton spin density in the extracellular space were found between healthyapples and those presenting internal disorders. The properties or factors that couldjustify the differences observed in these results such as the diffusion through magneticfield gradients, the magnetic susceptibility discontinuities or the physiologicalchanges during the development of this disorder were not investigated.

Chayaprasert and Stroshine (2005) carried out a study on the feasibility of globalmeasurements under on-line conditions. Experimental work was performed with








low-cost, low-field (0.13 T) equipment. The design of the NMR acquisition systemfor dynamic measurements was based on that proposed by Chen et al. (1996). Theauthors also claim that they have approached the concerns exposed by Hills andClark (2003) on the transferability of the NMR systems toward on-line conditions.Experiments were conducted on ‘Rome’ and ‘Red Delicious’ apples. Transverserelaxation time (T2) was computed for whole healthy and affected apples by meansof a CPMG sequence (908–1808 pulses with pulse interspacing of 600 ms). Becauseof the sample motion, the CPMG sequence could not be completed before the appleleft the sensitive area of the coil. Apparent decaying time T2 (AT2) was obtained aftercurve normalization, together with the sum of the areas of the normalized echo peaks(SUM). In this study both parameters AT2 and SUM were found lower for appleswith internal browning than those of the healthy fruit, even though the differencesbetween the relaxation curves of healthy and affected apples reduced as conveyorspeed increased.

To evaluate the effect of sample motion, CPMG signal acquisitions were testedat different belt speeds up to 250 mm=s at apples misalignments from �6 to þ6 mmwith respect to the coil center. The increasing conveyor belt speed induced adecreasing in the initial signal amplitude and a more rapid decay of the signal(Figure 5.3.10). The decreasing initial signal amplitude was attributed to the lowdegree of sample magnetization achieved during the shorter time inside the magnet.Normalization of different curves eliminated the variations in initial amplitudes(Figure 5.3.10). In addition, at increasing conveyor speed, samples get across thecoil in less time, so that in some of the experimental conditions the pulse sequence isnot fast enough to encode all data and thus the decaying curve is highly affected.Moreover, CPMG sequence does not compensate for the loss of phase coherencecaused by magnetic field inhomogeneities along the direction of the sample motion.Thus, AT2 decreases at higher conveyor speeds as larger distances are covered.Sample misalignments also exert an effect on the decay rate as consequence from the

1

0.8

0.6

0.4

Nor

mal

ized

am

plitu

de

0.2

00 1000

250 mm/s

150 mm/s

50 mm/s

0 mm/s

2000

Time [ms]

3000

FIGURE 5.3.10 Normalized signal amplitude of the echo train in CPMG sequence acquiredfrom an apple when conveyed at different speeds where increasing decaying rates are observedat increasing speeds. (Reprinted from Chayaprasert, W. and Stroshire, R., Postharvest Biol.Technol., 36, 291, 2005. With permission.)


different times for each sample to pass across the RF coil and as the field inhomo-geneities are not comparable.

The best classification performance was achieved at 50 mm=s by using the SUMparameter with classification error of 12% for ‘Rome’ apples and 0% for ‘RedDelicious’ apples. The extent of the damaged region was always higher than 40%of the whole tissue. Performance decreased to 20% of classification error at higherspeeds. Misalignment also caused a significant increase of the errors. As a conclu-sion, from this study, it is possible to detect apples with internal browning atconveyor belt speed below 100 mm=s, as long as the misalignment of the samplesis controlled. Recommended improvements to the system included a more robustconveyor system and modification of the magnet design to achieve a greater degreeof pre-magnetization. Authors also highlighted the need of determining the effects ofspacing between samples on the performance of the system under continuousoperating conditions.

5.3.4.4 Mealiness in Apples and Wooliness in Peaches

Mealiness is a negative attribute of sensory texture that combines the sensation of adesegregated tissue with the loss of crispiness and a lack of juiciness withoutvariation of the total water content in tissues. For mealy apples, these sensationsare the result of the weakness of the middle lamella that involves cells separationinstead of rupturing with releasing of aqueous content. Peach mealy textures are alsoknown as wooliness. In this case, the characteristic lack of juiciness derives from adifferent origin. Under disorder-inducing conditions, cell membrane permeability isaltered and pectin accumulates in cell walls and intercellular spaces. Those pectinsinteract with calcium, among other ions or molecules, and generate gel structures thatretain the water molecules. In addition, cell adhesion is reduced, which also means alack of crispiness. This physiological disorder is associated with inadequate over-extended cold storage.

Barreiro and colleagues (1999) assessed mealiness in apples through spatiallyresolved MRR as T2 maps corresponding to a centered tissue slice. Experiments wereconducted in a 4.7 T magnet with stationary samples. Several parameters wereextracted from the maps themselves such as the minimum, the average, and themaximum T2 values with their associated SDs. The minimum T2 value was the onlyparameter that showed significant differences between the non-mealy and the mealytexture stage of the fruit being shorter for affected apples. Differences were alsofound in the corresponding histograms. For mealy apples, histograms were skewedto shorter T2 times, whereas the distribution of the T2 values of sound apples wasnormal. In addition, the histogram of mealy apples showed a tail in the region of thehighest values. A new study (Barreiro et al. 2000) also based on spatially resolvedMRR reinforced previous results. In this case, more parameters were extracted fromthe T2 histograms and an analysis of variance revealed those with the highest effecton texture categories, that is, fresh, intermediate, and mealy fruits. Such parametersentered to a stepwise discriminant analysis to create classification functionsthat segregate between categories. An 87% of well-classified apples was achievedby combining the mode height-to-interquartile range ratio, the maximum T2 value,


the number of pixel s with T2 va lue below 35 ms, the upper quart ile, the modeheight, and T2 SD.

Mealiness (or wool iness) in pe aches was studied on a set of frui ts includingcrispy, intermed iate (non-c rispy but non-wool ly), woolly, and very wool ly peaches(Barreiro et al. 2000). Simi lar to apples, histogram s extra cted from the T2 mapsshowed a decrease in T2 values with disorder develo pment (Figure 5.3.11).

The lack of a pplicabilit y of spatiall y resol ved MRR ( T2 maps ) to on-lineinspectio n owing to the long acquis ition times redir ected subseq uent studies(Barreiro et al. 2002), which wer e focuss ed on rapid mealines s detection by non-spatially resol ved MR R. Previo us ac quisition of T2 maps (Figur e 5.3.12) wascomplemented using CPMG experiments with a 908–1808 pulse spacing of 200 msundertaken at 2.32 T on pieces of tissue extracted from apples presenting threelevels of disorder, that is, non-mealy apples, medium stage apples, and mealy apples.The T2 spectroscopic data showed three peaks. The dominant one, which alsoshowed the longest T2 value, was assigned to water in vacuole; the other twopeaks were associated with cell wall and cytoplasm, respectively. The main relax-ation time showed a significant decrease from 458.6 to 104.4 ms (repeatability of�21.6 ms) for increasing mealiness level. These results highlighted a good prospectfor on-line mealiness assessment.

In the same work, physical tissue properties related to tissue desegregation anddryness were analyzed with respect to T2 maps parameters. Linear positive corre-lation (r¼ 0.85) was found between tissue hardness and minimum value at T2 maps.Juiciness and SD of T2 values in the corresponding maps showed nonlinear negativecorrelation (r¼�0.80).

5.3.4.5 Internal Breakdown in Pears

Internal browning in pears is characterized by softening and browning of tissues anddevelopment of cavities. It is an important postharvest disorder that is observableonly at the end of the commercial chain since the external appearance of the fruit isnot altered, even when the characteristic brown colored tissue is widely spread fromthe core to the surrounding flesh. Many authors have identified the elevated CO2 anddecreased O2 levels in the air composition during CA storage as the primaryinfluence. Under such conditions, a decrease in antioxidant levels as well as inenergy availability is induced so that membrane maintenance and free radical controlare altered (Saquet et al. 2003; Veltman et al. 2003; Larrigaudière et al. 2004) in‘Conference’ pears and in ‘Blanquilla’ pears, leading to cell decompartmentation andbringing browning reactions. Other characteristics favoring the development of thedisorder include longer storage time along with overmaturity, heavy fruits, and hardtissue (Lammertyn et al. 2000) in ‘Conference’ pears. Thus, various names have beenapplied to call this disorder since it has been difficult to establish clear differences onthe basis of similar symptoms. Among these are internal breakdown, brown heart,brown core, and core breakdown.

Several authors have used NMR techniques (MRI and MRR) to study break-down development in ‘Bartlett’ and ‘Conference’ pears. Wang and Wang (1989)obtained SE images from ‘Bartlett’ pears with a 0.5 T NMR equipment. Bright


Very crispy

T2-maps T2-histograms

Num

. obs

erva

tions

Num

. obs

erva

tions

Num

. obs

erva

tions

Num

. obs

erva

tions

T2 [ms]

T2 [ms]

T2 [ms]

T2 [ms]

30

25

20

15

10

5

0

50.2

54.2

58.2

62.2

64.2

70.2

74.1

78.1

82.1

88.1

90.1

94.1 98 102

100

30

25

20

15

10

5

0

50.2

54.2

58.2

62.2

64.2

70.2

74.1

78.1

82.1

88.1

90.1

94.1 98 102

100

30

25

20

15

10

5

0

50.2

54.2

58.2

62.2

64.2

70.2

74.1

78.1

82.1

88.1

90.1

94.1 98 102

100

30

25

20

15

10

5

0

50.2

54.2

58.2

62.2

64.2

70.2

74.1

78.1

82.1

88.1

90.1

94.1 98 102

100

Non-crispybut non-woolly

Woolly

Very woolly

FIGURE 5.3.11 Examples of T2 maps of a very crispy, a non-crispy but not woolly, awoolly, and a very woolly peaches, and the corresponding histograms where horizontal axisrepresents the T2 values in milliseconds. (Reprinted from Barreiro, P., Ortiz, C., Ruiz-Altisent,M., et al., Magn. Resonance Imaging, 18, 1175, 2000. With permission.)

areas appeared in the MR images that corresponded to the affected region visuallydetected. The hyperintense signal was associated with an increase in the free watercontent. The presence of air spaces was revealed by dark pixel regions since PDdecreases drastically in these spaces.


250200

100

500

1 21 41 61 81 101

121

141

161

181

201

221

241

150

120100

6040

Num

ber

of p

ixel

sN

umbe

r of

pix

els

20

Gray level

Gray level

Non-mealy apple

Mealy apple

431087 μs ± 22946

104415 μs ± 5380

0

1 21 41 61 81 101

121

141

161

181

201

221

241

80

FIGURE 5.3.12 On the left, examples of MR images acquired from non-mealy and mealyapples. On the right, histograms from the corresponding T2 map of each apple with increasinggray level corresponding to increasing T2 values, and main T2 peak. (Adapted from Barreiro,P., Moya, A., Correa, E., et al., Appl. Magn. Resonance, 22, 387, 2002.)

Lammertyn and coworkers studied the spatial distribution (2003a) as well as thetime course (2003b) of the disorder in ‘Conference’ pears by means of MRI, spatiallyresolved MRR, and x-ray computer tomography. Series of T1-weighted SE MRimages were obtained, and PD and T1 maps were constructed.

In contrast to results of Wang and Wang (1989), affected brown tissue appearedin T1-weighted images as low-intensity signal regions, whereas healthy unaffectedtissue showed high-intensity signal. Two patterns of browning, radial and localdistributions, were observed.

The average PD values were found significantly higher for healthy tissue than forbrown tissue, which indicated the occurrence of dehydration for the latter. Inaddition, shorter T1 values were observed for the affected areas highlighting amore restricted motion, which involves higher pixel intensity on T1-weightedimages. Therefore, the net decrease in signal intensity for affected tissue pointed tothe more important effect of the PD reduction on signal generation.

A hypothesis was proposed to explain differences with the results of Wang andWang (1989). A higher free water level appears as cell membranes are affectedby disorder-inducing conditions. When the moisture transport is faster than thecellular decompartmentation rate, the PD of the brown tissue is lower than that ofthe healthy tissue. Under faster membrane disintegration, more mobile protons leadto a signal intensity increase. The latter is prone to occur under extremely high CO2

concentrations.The study on the time course of this disorder (Lammertyn et al. 2003b) revealed

that it does not grow spatially over time but only increases in severity and so in


(A) (B)

10 23 42

58 71 88

119 156 178

T2∗-weighted

PD-weighted

FIGURE 5.3.13 (A) T1-weighted SE MR images from a stationary ‘Conference’ pearacquired along 178 days storage under core breakdown-inducing conditions. The numbersindicate days after harvest. (Reprinted from Lammertyn, J., T. Dresselaers, P. Van Hecke,P. Jacsók, M. Wevers, and B.M. Nicolaï, Postharvest Biol. Technol., 29, 19–28, 2003b. Withpermission.) (B) Examples of T2*-weighted and PD-weighted FLASH MR images (703 msand 484 ms acquisition time, respectively) after motion correction obtained from healthy (left)and affected (right) ‘Blanquilla’ pears while conveyed at 54 mm=s. (Reprinted from Hernán-dez-Sánchez, N., Hills, B.P., Barreiro, P., and Marigheto, N., Postharvest Biol. Technol., 44,260, 2007. With permission.)

contrast magnitude (Figure 5.3.13A), and that cavities grow at the expense of thebrown tissue. The authors of the study tentatively elucidated that the cavitiesformation is a result of the moisture transport toward the fruit boundary, whichagrees with the decrease in PD and the subsequent image contrast obtained in theirwork on spatial distribution (Lammertyn et al. 2003a).

MRI offered higher sensitivity for detecting incipient browning than x-ray, sinceat the first stage of the disorder the cellular decompartmentation does not affect thedensity of the tissue but the mobility of the protons. In addition, a better contrastbetween affected and unaffected tissues was found for MRI inspection.

These earlier MRI and MRR studies were undertaken on stationary fruit usingslow imaging sequences. Lately in 2007, Hernández-Sánchez and the working groupperformed a study on ‘Blanquilla’ pears where macroscopic dynamic MRI experi-ments were complemented with optical microscopy and with non-resolved MRR andNMR diffusion analyses to provide insight into the effect of the disorder on themicroscopic tissue structure.

Optical microscopy confirmed that the cells lose their natural angular morph-ology and integrity in damaged tissues. This technique also showed that the whole


tissue loses compactness and coherence. Accumulation of vesicles containingbrown-colored compounds was also revealed by these microscopic images. Apreliminary study performed on pieces of affected and healthy tissue, and juicefrom a healthy pear (Hernández Sánchez, 2006) demonstrated the great effect ofthe tissue microstructure on the transverse relaxation rate (1=T2). CPMG sequenceswith varying pulse spacing pointed that such rate increases for altered tissue. Thisresult was related to the water diffusion through magnified local gradients as aconsequence of the magnetic susceptibility discontinuity across the phenolic vesiclesand water interfaces. This effect was enhanced for 7 T compared to 2.35 T, aslocal gradients are proportional to the applied magnetic field. For juice, the waterdiffusivity effect on T2 disappeared as tissue compartmentation is destroyed.

To obtain the probability density of different proton pools distinguished on thebasis of their T1 and T2 values, 2D (T1–T2) cross-correlation data were obtained byacquiring CPMG sequences at different inversion times, and performing 2D inverseLaplace transformation. Proton pool assignment for tissue characterization wasimplemented by performing (T2–D) correlation spectroscopy where the T2 of thedifferent water subcellular compartments were associated with water diffusion con-stants, D [m2=s].

T1–T2 correlation data showed two main water proton pools distinguishable forfresh tissue (Figure 5.3.14), which were associated with water in vacuole and waterin cytoplasm according to studies on parenchyma tissue of apple (Hills et al. 2004).These peaks merged when tissue was affected and a decrease in the T2 values was

10−4

10−3

10−2T2

[s]

T2

[s]

T1 [s]

10−1

100

101

10−4 10−3 10−2

C

10−1 100 10110−4

10−3

10−2

T1 [s]

10−1

100

101

10−4 10−3 10−2 10−1 100 101

V + C +sugar-CH

V +sugar-CH

FIGURE 5.3.14 T1–T2 correlation spectra obtained from healthy pear tissue (left) andflesh presenting internal browning (right) at 7 T with a 908–1808 pulse spacing of 4000 ms.Contour lines delimit the probability of protons populations with peaks merging in affectedtissue. C stands for water protons in cytoplasm, V stands for water protons in vacuole, andsugar-CH stands for protons in sugar molecules. (Reprinted from Hernández-Sánchez,N., Hills, B.P., Barreiro, P., and Marigheto, N., Postharvest Biol. Technol., 44, 260, 2007.With permission.)


observe d (Figur e 5.3.14). Peak merging highl ights the loss of mem brane inte grityand the enhancem ent of the diffusion exchange , which was con fi rmed by the higherdiffusion coef ficient compu ted by the T2–D correl ation spectroscop y for the vacuol ecompartm ent in affect ed tissue. The observ ed T2 shortening was explained by thewater diffusion through stro ng internal field gradients arising from suscep tibilitydisconti nuities wi thin the affect ed tissue.

Such microscop ic informat ion was used to direc t the macr oscopic study . PD- andT2*-weight ed images were obtai ned with an in-plane resol ution of 0.88 mm 2 perpixel (8.8 mm 3 per pixel volume resol ution) using a 4.7 T magne t. The acquisition of50% of the phase- encodi ng steps and the subsequ ent zero- fi lling yiel ded totalacquisition times of 484 an d 703 ms, respective ly. Cor onal ima ges wer e acq uiredfrom the equato rial slice of intact frui ts conveyed at 54 mm=s and phase shift,induced by sample mot ion, was correct ed for image recons truction (Figur e5.3.13B). Discrimi nant funct ions were obtained by selec ting featu res, which wereautomati cally extra cted from the image hist ograms, among those show ing signi fican tdifferences between healthy and affected pears by ANO VA analys is. The achiev edcorrect classi ficatio ns were 98.4% for the form er and 91% for the latter pears whenusing PD-wei ghted image s; and 98% and 86% when using T2*-weight ed images.The final perfor mance rema ins to be vali dated at low-m agnetic field stren gths andhigher conveyor speeds .

5.3.4.6 Freeze Injury in Citrus

Freezin g injury in oranges may appea r whenev er a tem perature below the freezingpoint of the tissues is reached along the prehar vest period. Injured fruits usuallyremain on the tree without any exter nal symp toms. For inju red frui t, the juic e sacsdry out as the ice crystals burst the mem branes a nd the cell walls. Also, the presen ceof wat er-soaking areas on the segment membranes becom es apparen t for frozenfruits. Dehy drated tissues collapse at a severe stage, leading to the deve lopment ofhollows between and within the segments. The occurr ence of both dehydra tion andhollows in da maged fruits converts MRI as very suit able techni que for freezinginjury detection (Hernández Sánchez, 2006).

‘Val encia ’ oranges wer e obtained after intense freezing condit ions and measuredin on-li ne MRI experiment s using an NMR spectrome ter of 4.7 T (He rnández -Sánchez et al. 2004). The se samp les along wi th nonexpos ed oranges wer e stabi lizedat room temp erature. Then , oranges wer e placed in the convey or belt with theirstem –calyx axis along the z direc tion, and axial FLASH images with an in-planeresoluti on of 0.88 mm 2 per pixel (8.8 mm 3 per pixel volume resoluti on) obtai ned at0, 50, and 100 mm=s belt speeds . To reduce the total acquis ition time, half-F ouriergradient echo acquisitions with zero filling before image recons truct ion were used toobtain sub-second scan times (780 ms). The loss of the image quality derived fromlower k-space coverage (25%) did not compensate the significant reduction of theacquisition time (390 ms).

Affected tissue appeared as a region of hypointense signal, with differentintensity levels according to the damage severity grade, while bright pixels relatedto non-affecte d tis sue (Figure 5.3.15). Hypo intense signa l is derived from the


Stationary MR image Dynamic MR image

FIGURE 5.3.15 Example of axial FLASH images (780 ms acquisition time) obtained from afreeze injured orange while stationary (in the middle) and when conveyed at 54 mm=s (on theright). On the left, RGB image captured with a digital camera approximately correspondingto the center of the imaged FOV. (Adapted from Hernández-Sánchez, N., Barreiro,P., Ruiz-Altisent, M., Ruiz-Cabello, J., and Fernández-Valle, M.E., Appl. Magn. Resonance,26, 431, 2004.)

reduction in PD as juice content decreases and from T2* reduction caused by thelocal field gradients at the interfaces between juicy and dehydrated structures. Thecentral axis also showed hypointense signal as the mobility of the water is lower inthis solid-like structure.

For the automatic analysis, two threshold segmentations were performed: one toseparate the entire region of interest including healthy and affected tissue areas andthe other to calculate the area of the healthy regions. This segmentation procedureuses an iterative process that is repeated until the number of pixels comprising eachregion reaches convergence. The signal hypointense fruit region was computed bysubtracting those segmented areas. Thresholds were addressed at each belt speedfrom undamaged fruits as the ratio between the signal hypointense region and theregion of interest. The threshold value increased from 10% under static conditionsto 20% and to 30% at 50 and 100 mm=s belt speed, respectively. The lack ofconsistency was probably a consequence of the image blurring induced by theexcitation of consecutive slices driven by the axial location of the FOV togetherwith the sample motion. In fact, the resultant images were a rough projection of theentire orange volume. Authors indicated the major need of blurring reduction,although the use of coronal or sagittal images with motion correction procedureswas not evaluated.

Gambhir and coworkers (2005) performed an MRR study on the effect offreezing (�78C) and chilling (58C) temperatures on the overall proton transverserelaxation time (T2) of peel and juice sacs in ‘Navel’ oranges. Oranges were exposedto temperature treatment for 20 h. Samples warmed to room temperature and thenwere peeled. Measurements were conducted on a 0.235 T equipment. The exposureto chilling or freezing temperature did not affect the T2 values of peel suggesting thatthe peel did not freeze under these conditions, whereas at �208C the T2 reduceddrastically. Freezing temperature caused an appreciable decrease in the T2 values offlesh segments, which was associated with damage in the juice sacs membrane andsubsequent leakage of juice.



5.3.4.7 Seed Identi fi cation in Citrus

One of the main concern s of citrus produce rs is the presen ce of seeds wi thin orangesand mandarins as it greatly devalu es the product quality. The acqu isition of MRimages wher e seeds are distingu ishab le from juic e segm ents is one of the mostpromisi ng procedu res to identify seed-contai ning frui t. Seeds possess a solid-likestructure that confers noti ceable diff erences from the juicy pulp, which involve astraight forward c ontrast manag ement in convent ional NM R equipment s, and subse-quent imag e proces sing. The chall enge is the achiev ement of pract ical image contr astby wor king procedu res transferab le to the indus trial environ ment.

Blasco and colleagues (2003) evalua ted the feasi bility of obtain ing enough MRIcontrast to detect seeds in stat ionary oranges by using three types of sequenc es, thatis, SE, half SE, and gradi ent echo, with varyi ng TE, TR, and FOV . MR ima ges wereacquired with a medi cal NM R equipment of 0. 2 T magne tic field strength. Threeslices perpend icul ar to the stem -calyx axis of the frui t were ima ged per sample.Among the analyz ed sequenc es, SE with TR of 50 ms and TE o f 18 ms provided thebest discr iminato ry resul ts. An algorithm de veloped to detect the seeds within theimage achiev ed 100% of succes s. Howe ver, the acquis ition time could not be setshorter than 7 s.

Hernández and coll eagues (2005) used a faster MR I sequenc e, a gradient-e choFLASH with TR of 12.2 ms, TE of 3.8 ms, a nd flip angle 10 8 to detect seed-c ontainingoranges when conveyed at 0, 54, and 90 mm=s throu gh the 4.7 T magne tic fieldstrength magne t of an NMR research equipment . Ora nges wer e placed in the conveyorbelt with their stem-caly x axis along the motion direction a nd axial images wereacquired with an in-plane resol ution of 0.88 mm 2 per pixel (8.8 mm 3 per pixel volumeresoluti on). The regio n contai ning the seeds and centr al axis tissues appeared darkerthan the flesh pixels. As could be ex pected, the higher the belt speed the higher theblurring artifact s caused by the superimpo sition of signa l arising from adjace nt slices.Neverthel ess, the contr ast was sti ll enough to allow a reliable a utomatic segm entationof the signa l hypoin tense region and featu re extractio n when samp les were conveyedat 50 mm=s belt speed with the acquis ition of the 50% of k -space lines (zero filling wasapplied before recons truction) . At higher speeds , internal structure s were compl etelymasked. An unsuper vised analysis on featu res extracted from the segm ented regionsrevealed that seedless oranges and oranges with one seed were grouped into a singlecluster. Taking into account that the discrimination between these two categories iscritical for commercial purposes, it was concluded that on-line axial images could notbe used for seed identification at current acquisition times (780 ms). In the same work(Hernández et al. 2005), coronal images from the equatorial slice of moving lemonswere obtained, and the performance of a low time-consuming motion correctionprocedu re was evalua ted (Figur e 5.3.16).

These promising results encouraged subsequent studies on mandarins (Hernández-Sánchez et al. 2006). Here, the NMR equipment also worked at 4.7 T and FLASHsequence was used. The imaging parameters were TR of 11 ms, TE of 3.8 ms, 108 flipangle, 50% of k-space lines, and zero filling, with a total acquisition time of 703 ms.Outstanding improvement in image quality was obtained since the features extractedfrom the segmented areas in motion-corrected images were not significantly different


MR image beforemotion correction

MR image aftermotion correction

FIGURE 5.3.16 Example of motion-induced artifacts correction in a coronal MR imageacquired during sample motion at 54 mm=s. On the left, RGB image captured with a digitalcamera approximately corresponding to the center of the imaged FOV; in the middle coronal,MR image reconstructed without artifact correction; on the right, same MR image recon-structed after artifact correction.

from those extracted from the static images (Figure 5.3.17). Those showing significantdifferences between seed-containing and seedless mandarins by ANOVA analysisentered a discriminant analysis. Finally, the perimeter of the region containing the

Stationary

Dynamic

FIGURE 5.3.17 Examples of both seedless (left) and seed containing (right) mandarins.Upper line corresponds to RGB images, middle line to static FLASH MR images, and bottomline to motion-corrected MR FLASH images acquired at 54 mm=s conveyor belt speed.(Reprinted from Hernández-Sánchez, N., Barreiro, P., and Ruiz-Cabello, J., BiosystemsEng., 95, 529, 2006. With permission.)



seeds and the central axis, and the maximum distance between the perimeter and thegravity center of this region were selected for classification. The number of samplescorrectly classified varied from 88.9% under static to 92.5% under dynamic conditionsfor seedless mandarins, and from 86.7% to 79.5% for fruits with seeds. A mainconclusion of this study was that it is feasible to perform a straightforward export ofmodels developed under static conditions toward on-line conditions.

Faster MRI sequences and more efficient segmentation procedures have beenevaluated recently by Barreiro and colleagues (2007) with stationary mandarins.Two different types of fast MRI sequences were investigated using a 4.7 T magnet: agradient-echo sequence (FLASH sequence) with 484 ms acquisition time and aspiral–radial sequence (COMSPIRA sequence) with 240 ms. Three segmentationtechniques were applied for image postprocessing, that is, region-based, 1D histo-gram variance, and 2D histogram variance. These studies have demonstrated that thelatter option provides the most promising results. Image features extracted from thesignal hypointense region including perimeter, compactness, maximum distance tothe gravity center, and aspect ratio are employed in a linear classification function, bywhich the seed identification can be achieved with 100% accuracy using spiral-radial[T2] sequence and 98.7% accuracy with gradient-echo images. These results need tobe validated with a higher number of samples and using a low field magnet.

5.3.5 CONCLUDING REMARKS

We presented the technical background and survey of modern applications of NMRin fruits and vegetables. By analyzing the enormous available literature, we tried toprovide an overview that could be interesting to those entering or to those who arecurrently applying other techniques in similar areas of agriculture and food. The highcost of commercial NMR instruments has probably prevented many authors to useNMR in their investigations. This is probably the reason why its use has notextended further. Being critical with the NMR work presented along these years,we can say that it has not yet gone beyond the consideration of being an interestingresearch tool. To go further with NMR applications, we should pay more attention toeconomy and efficiency, and break the vicious cycle that drives us to use only high-quality instruments, mainly designed to medical applications, whose quality stand-ards logically are much more severe and whose results cannot in many cases betransferred to a final non-experience user working in a completely different envir-onment such as that related to agricultural products. Joint efforts are worthy in viewof the expansion of the NMR technique.

The basics of the NMR confer its exceptional capabilities to inspect internalquality parameters and to provide valuable insight on physiological processes as it issensitive to the content, chemical environment, mobility, and diffusion among otherphenomena, related to aqueous protons. These phenomena are influenced by cellcompartmentation and tissue microstructure so that the NMR signal also acts assensor of the tissue integrity.

The techniques derived from the NMR such as NMR relaxometry, NMR spec-troscopy, and MRI greatly broaden the range of applications devoted to the inspec-tion of food products and, particularly, of fruits and vegetables.


The different relaxation times obtained by means of MRR, longitudinal relax-ation time T1, and transverse relaxation time T2 (including effective T2*) are sourcesof contrast that carry diverse underlying information for different types of tissue. T1distribution reveals differences between molecular motions within the tissue, whichare related to their size and complexity, and the viscosity of the surroundings. As forT2, the tumbling motions of the molecules and the chemical exchange betweenproton pools of different metabolites give rise to relaxation contrast. The diffusionthrough local magnetic field inhomogeneities induced by susceptibility variations atdifferent interfaces and microstructure arrangements is also responsible for T2differences. The cell morphology and size, the membranes permeability, andhence, the membranes integrity affect such diffusion.

The MRS provides both qualitative and quantitative information on the compos-ition of the sample with regard to a specific nucleus. Variations in electronic andchemical environment induce the groups of chemically equivalent nuclei of eachmolecule to precess at different frequencies. The qualitative information is providedby the different chemical shifts at which the resolved peaks are located in the NMRspectrum. The quantitative information is derived from the area of each peak.

MRI provides a picture that contains combined spectroscopy and relaxometryinformation both spatially resolved. An MR image is basically an indication of thePD contained in the sample enriched with the relaxation information. The manage-ment of the image acquisition parameters allows different weightings. In PD-weighted images, those regions with higher number of nuclei appear brighter, forT1-weighting the bright pixels correspond to tissues with short T1, whereas for T2-weighted images the longer T2 components present higher intensity signal.

The studies performed up to date are a demonstration of the potential of thesetechniques for the internal quality monitoring. Continuous technological advances inNMR hardware and computers allow more efficient explorations and are allies forthe achievement of new correlations between quality aspects and NMR features aswell as for the confirmation of those already found. However, there is additionalintensive work that must be faced to optimize the NMR systems and the signalacquisition procedures, so that the transference of the acquired knowledge to com-mercial equipments is more realistic.

The implementation of the NMR techniques (MRR, MRS, and MRI) demandsthe use of appropriate technology at low-magnetic field strengths, efficient pulsesequences with ultrafast acquisition times (below 1 s per fruit), and the inspectionunder motion conditions to reach the typical conveyor velocities (up to 2 m=s). Asaforementioned, one of the major hindrances for the transferability of the currentapplications to the horticultural industry is the high cost of the equipment used forresearch works, which is a consequence of the high field strength magnets and thehigh-performance hardware required for the application of some MRI sequences.

When comparing high- and low-field MRI applications, limits for the use at low-magnetic field strength mainly derive from the decrease in SNR (owing to the poorersample polarization) for a similar pulse sequence, acquisition parameters, and coil.Other factors, such as relaxation times, magnetic susceptibility differences, and theloss of contrast arising from local field gradients (as local gradients decrease at lowermagnetic field strength), need to be evaluated individually for each application, since


the changes in these properties can be decisive for inspection performance. Hills andClark (2003) proposed to overcome the loss of SNR at low field by using aprepolarizing unit. These authors also highlighted the field homogeneity withinlarge volumes and the compatibility with commercial graders as major subjects tobe considered.

For fast sample inspection with low hardware demand MRS, nonspatiallyresolved CPMG-T2 measurements and 1D MRI provide the most efficient results.As for 2D MRI, high-quality images have been obtained with very short acquisitiontimes, and motion-induced artifacts have been successfully corrected for coronalFOVs. However, it is a very demanding technique and validation of the results atlow-magnetic field remains to be carried out. Nevertheless, the optimization anddevelopment of single-shot sequences along with alternative k-space sampling stra-tegies bring optimistic expectations. In contrast, spatially resolved MRR experiments(e.g., relaxation maps) are often highly time consuming, although the providedinformation on inherent tissue characteristic is critical for the optimization of themore transferable experimental procedures.

With regard to the inspection under continuous motion, it seems to be a majorrequirement to reach the current conveyor velocities. Besides, the possible tumblingof the samples induced by the conveyor start and stop would bring artifacts thatcould spoil the measurement reliability. Additional time for sample stabilizationwould be needed. However, the industry could compromise the reduction in inspec-tion rate as long as final results conform to significant improvements in qualitystandards.

The main applications that have been lately evaluated under on-line conditionsare maturity in avocados, pit detection in cherries and olives, internal browning inapples, internal breakdown in pears, freeze injury in citrus, and seed in citrus, all ofthem related to internal quality aspects, which are not as easily detected with othertechniques. Up to date, these applications have not been implemented by industry.

Collaborative works require a closer interaction between researchers and thefood industry, as well as the implication of the NMR manufacturers to converge onrealistic and reliable monitoring systems that exploit the great potential of the NMRtechniques. There are some research groups that have already started, as those led byMJ McCarthy (University of California, United States) and BP Hills (Institute ofFood Research, United Kingdom), who are better situated to materialize the encour-aging expectations in the near future.

REFERENCES

Barreiro, P., J. Ruiz-Cabello, M.E. Fernández-Valle, C. Ortiz, and M. Ruiz-Altisent. 1999.Mealiness assessment in apples using MRI techniques. Magnetic Resonance Imaging17: 275–281.

Barreiro, P., C. Ortiz, M. Ruiz-Altisent, J. Ruiz-Cabello, M.E. Fernandez-Valle, I. Recasens,M. Asensio. 2000. Mealiness assessment in apples and peaches using MRI techniques.Magnetic Resonance Imaging 18: 1175–1181.

Barreiro, P., A. Moya, E. Correa, M. Ruiz-Altisent, M. Fernández-Valle, A. Peirs, K.M.Wright, and B.P. Hills. 2002. Prospects for the rapid detection of mealiness in applesby nondestructive NMR relaxometry. Applied Magnetic Resonance 22: 387–400.


Barreiro, P., C. Zheng, D.W. Sun, N. Hernández-Sánchez, J.M. Pérez-Sánchez, andJ. Ruiz-Cabello. 2007. Non-destructive seed detection in mandarins: Comparison ofautomatic threshold methods in FLASH and COMSPIRA MRIs. Postharvest Biologyand Technology. doi:10.1016=j.postharvbio.2007.07.008.

Bernstein, M.A., K.F. King, and X.J. Zhou. 2004. Handbook of MRI Pulse Sequences.Elsevier Academic Press, United States of America.

Blasco, J., M.C. Alamar, and E. Moltó. 2003. Detección no destructiva de semillas enmandarinas mediante resonancia magnética. Resúmenes del II Congreso Nacional deAgroingeniería, Córdoba, Spain, 439–440.

Bracewell, R.N. 2000. The Fourier Transform and its applications, 3rd ed. McGraw-HillHigher Education, Casson T, CA.

Brescia, M.A., T. Pugliese, E. Hardy, and A. Sacco. 2007. Compositional and structuralinvestigations of ripening of table olives, Bella della Daunia, by means of traditionaland magnetic resonance imaging analyses. Food Chemistry 105: 400–404.

Butz, P., C. Hofmann, andB. Tauscher. 2005. Recent developments in noninvasive techniques forfresh fruit and vegetable internal quality analysis. Journal of Food Science 70: 131–141.

Chayaprasert, W. and R. Stroshire. 2005. Rapid sensing of internal browning in whole applesusing low-cost, low-field proton magnetic resonance sensor. Postharvest, Biology andTechnology 36: 291–301.

Chen, P., M.J. McCarthy, S.M. Kim, and B. Zion. 1996. Development of a high-speed NMRtechnique for sensing maturity of avocados. Transactions of the American Society ofAgricultural Engineers 39: 2205–2209.

Clark, C.J., P.D. Hockings, D.C. Joyce, and R.A. Mazzuco. 1997. Application of magneticresonance imaging to pre- and post-harvest studies of fruits and vegetables. PostharvestBiology and Technology 11: 1–21.

Clark, C.J. and D.M. Burmeister. 1999. Magnetic resonance imaging of browning develop-ment in ‘Braeburn’ apple during controlled-atmosphere storage under high CO2. Horti-cultural Science 34: 915–919.

Delpuech, J. 1995. Dynamics of Solutions and Fluids Mixtures by NMR. John Wiley & Sons,Chichester, United Kingdom.

Gambhir, P.N., Y.J. Choi, D.C. Slaughter, J.F. Thompson, and M.J. McCarthy. 2005. Protonspin–spin relaxation time of peel and flesh of navel orange varieties exposed to freezingtemperature. Journal of the Science of Food and Agriculture 85: 2482–2486.

Goñi, O., M. Muñoz, J. Ruiz-Cabello, M.I. Escribano, and C. Merodio. 2007. Changes inwater status of cherimoya fruit during ripening. Postharvest Biology and Technology 45:147–150.

González, J.J., R.C. Valle, S. Bobroff, W. Biasi, E.J. Mitcham, and M.J. MacCarthy. 2001.Detection and monitoring of internal browning development in ‘Fuji’ apples using MRI.Postharvest Biology and Technology 22: 179–188.

Hernández-Sánchez, N., P. Barreiro, M. Ruiz-Altisent, J. Ruiz-Cabello, and M.E. Fernández-Valle. 2004. Detection of freeze injury in oranges by magnetic resonance imaging ofmoving samples. Applied Magnetic Resonance 26: 431–445.

Hernández, N., P. Barreiro, M. Ruiz-Altisent, J. Ruiz-Cabello, and M.E. Fernández-Valle.2005. Detection of seeds in citrus using magnetic resonance imaging under motionconditions and improvement with motion correction. Concepts in Magnetic ResonancePart B: Magnetic Resonance Engineering 26B: 81–92.

Hernández Sánchez, N. 2006. Development of on-line NMR applications for the evaluation offruit internal quality. Dissertation at the Polytechnic University ofMadrid, Madrid, Spain.

Hernández-Sánchez, N., P. Barreiro, and J. Ruiz-Cabello. 2006. On-line identification of seedsin mandarins with magnetic resonance imaging. Biosystems Engineering 95: 529–536.

Hernández-Sánchez, N., B.P. Hills, P. Barreiro, and N. Marigheto. 2007. An NMR study oninternal browning in pears. Postharvest Biology and Technology 44: 260–270.


Hills, B.P. and B. Remigereau. 1997. NMR studies of changes in subcellular water compart-mentation in parenchyma apple tissue during drying and freezing. International Journalof Food Science and Technology 32: 51–61.

Hills, B.P. 1998. Magnetic Resonance Imaging in Food Science. John Wiley & Sons, NewYork.

Hills, B.P. and C.J. Clark. 2003. Quality assessment of horticultural products by NMR. AnnualReports on NMR Spectroscopy 50: 75–120.

Hills, B., S. Benamira, N. Marigheto, and K. Wright. 2004. T1-T2 correlation analysis ofcomplex foods. Applied Magnetic Resonance 26: 543–560.

Jung, K.H., R. Stroshine, P. Cornillon, and P.M. Hirst. 1998. Low field proton magneticresonance sensing of water core and internal browning in whole apples. AmericanSociety of Agricultural Engineers Conference Paper 98-6020.

Keener, K.M., R.L. Stroshine, and J.A. Nyenhuis. 1997. Proton magnetic resonance measure-ment of self-diffusion coefficient of water in sucrose solutions, citric acid solutions, fruitjuices, and apple tissue. Transactions of the American Society of Agricultural Engineers40: 1633–1641.

Kim, S.M., P. Chen, M.J. McCarthy, and B. Zion. 1999. Fruit internal quality evaluation usingon-line Nuclear Magnetic Resonance Sensors. Journal of Agricultural EngineeringResearch 74: 293–301.

Korin, H.W., F. Farzaneh, R.C. Wright, and S.J. Riederer. 1989. Compensation for effects oflinear motion in MR imaging. Magnetic Resonance in Medicine 12: 99–133.

Lammertyn, J., M. Aerts, B.E. Verlinden, W. Schotsmans, and B.M. Nicolaï. 2000. Logisticregression analysis of factors influencing core breakdown in ‘Conference’ pears. Post-harvest Biology and Technology 20: 25–37.

Lammertyn, J., T. Dresselaers, P. Van Hecke, P. Jacsók, M. Wevers, and B.M. Nicolaï. 2003a.MRI and X-ray CT study of spatial distribution of core breakdown in ‘Conference’pears. Magnetic Resonance Imaging 21: 805–815.

Lammertyn, J., T. Dresselaers, P. Van Hecke, P. Jacsók, M. Wevers, and B.M. Nicolaï. 2003b.Analysis of the time course of core breakdown in ‘Conference’ pears by means of MRIand X-ray CT. Postharvest Biology and Technology 29: 19–28.

Larrigaudière, C., I. Lentheric, J. Puy, and E. Pintó. 2004. Biochemical characterisation of corebrowning and brown heart disorders in pear by multivariate analysis. PostharvestBiology and Technology 31: 29–39.

Létal, J., D. Jirák, L. Šuderlová, and M. Hájek. 2003. MRI ‘‘texture’’ analysis of MR images ofapples during ripening and storage. Lebensmittel-Wissenschaft und-Technologie 36:719–727.

Marigheto, N., S. Duarte, and B.P. Hills. 2005. An NMR relaxation study of avocado quality.Applied Magnetic Resonance 29: 687–701.

McCarthy, M.J. 1994. Magnetic Resonance Imaging in Foods. Chapman and Hall.Parikh, A.M. 1992. Magnetic Resonance Imaging Techniques. Elsevier, New York.Pathaveerat, S., M.J. McCarthy, and P.P. Chen. 2001. Spatial distribution of avocado com-

position: Implications for on-line sorting by NMR spectroscopy. VI InternationalSymposium on Fruit, Nut, and Vegetable Production Engineering. Potsdam (Germany),Conference Proceedings, 645–649.

Pérez-Sánchez, J.M., I. Rodríguez, R. Pérez De Alejo, M. Cortijo, and J. Ruiz-Cabello. 2006.COMSPIRA3D: A combined approach to radial and spiral 3D MRI. Concepts inMagnetic Resonance Part B: Magnetic Resonance Engineering 29B: 107–160.

Raffo, A., R. Gianferri, R. Barbieri, and E. Brosio. 2005. Ripening of banana fruit monitors bywater relaxation and diffusion 1H-NMR measurements. Food Chemistry 89: 149–158.

Rodríguez, I., R. Pérez de Alejo, J. Cortijo, and J. Ruiz-Cabello. 2004. COMSPIRA: Acommon approach to spiral and radial MRI. Concepts in Magnetic Resonance Part B:Magnetic Resonance Engineering 20B: 40–44.


Saquet, A.A., J. Streif, and F. Bangerth. 2003. Energy metabolism and membrane lipidalterations in relation to brown heart development in ‘Conference’ pears during delayedcontrolled atmosphere storage. Postharvest Biology and Technology 30:123–132.

Thybo, A.K., P.M. Szczypinski, A.H. Karlsson, S. Donstrup, H.S. Stodkilde-Jorgensen, andH.J. Andersen. 2004. Prediction of sensory texture quality attributes of cooked potatoesby NMR-imaging (MRI) of raw potatoes in combination with different image analysismethods. Journal of Food Engineering 61: 91–100.

Tu, S.S., J.C. Young, M.J. McCarthy, and K.L. McCarthy. 2007. Tomato quality evaluationby peak force and NMR spin-spin relaxation time. Postharvest Biology and Technology44: 157–164.

Veltman, R.H., I. Lenthéric, L.H.W. Van der Plas, and H.W. Peppelenbos. 2003. Internalbrowning in pear fruit (Pyrus communis L. cv Conference) may be a result of a limitedavailability of energy and antioxidants. Postharvest Biology and Technology 28:295–302.

Vlaardingerbroek, M.T. and J.A. den Boer. 1996. Magnetic Resonance Imaging. Springer-Verlag, Berlin=Heidelberg, Germany.

Wang, C.Y. and P.C. Wang. 1989. Nondestructive detection of core breakdown in ‘Barlett’pears with nuclear magnetic resonance imaging. Horticultural Science 24: 106–109.

Woodward P. 1995. Contrast. In MRI for Technologists. McGraw-Hill Education, USA.Zion, B., M.J. McCarthy, and P. Chen. 1994. Real-time detection of pits in processed cherries

by magnetic resonance projections. Lebensmittel-Wisenschaft und -Technologie 27:457–462.

Zion, B., S.M. Kim, M.J. McCarthy, and P.J. Chen. 1997. Detection of pits in olives undermotion by nuclear magnetic resonance. Journal of the Science of Food and Agriculture75: 496–502.


5)spectroscopic methods

Documents