Introduction

Sources of Intrinsic Optical Contrast in BreastTissue

Principles of Optical Mammography

Continuous-wave Approaches: DynamicMeasurements and Spectral Information

Time-resolved Approaches

Interpretation of Optical Mammograms

Prospects of Optical Mammography

References

Introduction

White-light transillumination was introduced into medi-cine in the early 1800s, but only in the late 1920s was itapplied for breast imaging and visualization of breastlesions. In a darkened room, the breast was illuminated withpowerful white light, and the transmitted image wasobserved directly by eye on the other side of the pendantbreast, looking for shadows that might reveal the presence ofa pathologic condition. The procedure was simple and inex-pensive, and it proved useful to identify hematomas andliquid cysts. However, it did not reach wide applicability,

because it did not enable discrimination between malignantand benign solid lesions, and some physicians had problemswith image interpretation. In the 1980s, new systematicattempts were made to apply optical techniques to breastimaging, and led to a method of transillumination calleddiaphanography, or lightscanning. A tungsten filamentlamp, filtered to select red and near-infrared (NIR) light,was used for illumination, typically through a fiber-optichand-held illuminator applied to the breast, and the breastshadow was recorded on an infrared-sensitive film or with avideo camera connected to a black-and-white monitor andvideotape recorder. In a diaphanography examination, lightwas diffused throughout the breast and randomly scattered.Opaque lesions formed shadows on the surface of the breastthat acted as a screen. The deeper is the lesion, the greaterthe distance from the screen, and the less the contrast. Thisimplies an inherent limitation of the technique, since smalllesions will only appear with high contrast if they are not toofar from the surface. Typically, high absorption, showing asa dark shadow, was regarded as the most specific sign ofabnormality, but also asymmetry between the two breastsand abnormal vasculature (again visualized as dark shad-ows) were considered in the interpretation of the images.

Initial studies, performed on a small series of selectedpatients, suggested the possibility of successfully detectingsolid breast tumors (Ohlsson et al., 1984). Those positive

16

Optical MammographySergio Fantini and Paola Taroni

▲ ▲Cancer Imaging Lung and Breast Carcinomas

Copyright © 2007 by Elsevier, Inc.All rights of reproduction in any form reserved.449

▲ ▲

Ch4.16-P370468.qxd 3/15/07 5:51 PM Page 449

outcomes fostered the research on diaphanography: com-mercial instruments became available and systematic trialswere carried out, including blind studies performed prospec-tively in a screening population to compare diaphanographywith X-ray mammography. Contrasting results wereobtained. In some cases, optical methods compared favor-ably with X-rays (Wallberg et al., 1985), while other studieswere negative about the diagnostic potential of diaphanogra-phy (Monsees et al., 1987). The latter studies highlighted ahigh number of “equivocal” or false-positive findings,which could only partially be retrospectively explained withtechnical limitations such as limited field of view, insuffi-cient illumination level, or inadequate positioning of thelight source. The sensitivity was also significantly lowerthan with mammography (about 60 to 70% versus 90 to98%), especially when small lesions were considered.Moreover, previous knowledge of the lesion locationseemed to be a critical factor, leading to much better resultsin retrospective analysis of data that were originallyacquired and first analyzed prospectively (e.g., an increasein sensitivity from 76 to 94% was achieved by Bartrum andCrow, 1984).

Despite the mixed results, even the researchers whowould not recommend diaphanography as a screeningmethod stressed its advantages in terms of noninvasiveness,good patient acceptance, quick and easy examination, andcost-effective instrumentation. They also suggested poten-tial adjunct roles for diaphanography, like improving thepositive yield of biopsies among patients recommended forsurgery or following up equivocal lesions at frequent inter-vals. However, the negative results in the diagnostic dis-crimination of solid breast tumors discouraged furtherdevelopment of diaphanography.

A first reason for the failures of optical methods in theirearly application to the detection and classification of breastlesions is the insufficient training and experience of theinvestigators performing the examination and, more impor-tantly, interpreting the images. But a second and more con-ceptual limitation of diaphanography was associated withthe instrumentation used and the lack of a physical model toquantitatively describe light propagation in breast tissue. Asa result, diaphanography could not address crucial chal-lenges in optical imaging associated with the diffusivenature of light propagation in breast tissue, the strong atten-uation of light, and the sensitivity to breast boundaries.Furthermore, diaphanography did not take full advantage ofthe diagnostic value of the spectral information that can beobtained with optical mammography. More recently, startingin the 1990s, technical improvements on both instrumentalaspects and theoretical modeling have suggested possibili-ties to better exploit the optical properties of healthy and dis-eased breast for diagnostic purposes, and have opened newopportunities to the application of optical methods for breastcancer detection.

Sources of Intrinsic Optical Contrast in Breast Tissue

A practical problem associated with optical techniques isthe strong light attenuation in biological tissues, which pre-vents whole-body examinations. Even in more favorablecases, dealing with a few centimeters of soft tissue, as in thecase of breast imaging, the attenuation of ultraviolet orblue/green light is too strong for any practical application. Asa result, red and near-infrared (NIR) light is always used inpractice for its relatively low attenuation. The red-NIR spec-tral region, say from 600 to 1000 nm, is sometimes called theoptical diagnostic window. As described above, diaphanogra-phy relied on the direct visualization of a shadow resultingfrom a localized increase in light attenuation. Two opticalphenomena contribute to the attenuation of light: absorptionand scattering. Absorption annihilates the incoming photons,thus reducing the intensity of the transmitted light. Scatteringevents are essentially changes in the direction of propagationof photons. Even though photons continue to propagate inthe medium, their change in direction contributes to theattenuation of the light transmitted through tissue along theoriginal direction of propagation. To fully understand thediagnostic potential of optical data, one needs to consider theorigin of both absorption and scattering phenomena.

Different substances typically absorb at different wave-lengths. Thus, at least in principle, the measurement of theabsorption spectrum of a medium allows for identificationof its constituents. Moreover, the higher is the amount of aspecific constituent, the stronger is its relative contributionto the overall absorption of tissue. Thus, the absorptionproperties of tissue also provide information on the concen-tration of its various constituents. Early studies have shownthat hemoglobin contributes markedly to breast absorptionat red and NIR wavelengths. This can have important diag-nostic implications, since areas of high vascularization canbe readily identified. Moreover, the two forms of hemoglo-bin (oxy-hemoglobin and deoxy-hemoglobin) have differentabsorption spectra. Therefore, changes in oxygen saturationlevels can also be optically measured. Water and lipids showcharacteristic strong absorption peaks in the NIR. The bal-ance of water and lipids in breast tissue depends on factorslike age and hormonal status. Furthermore, the water andlipid content of pathologic lesions is likely to differ fromthat of healthy tissue, making its assessment potentially use-ful for diagnostic purposes. From an experimental point ofview, to derive information on tissue composition, measure-ments must be performed at several wavelengths. In fact, theabsorption coefficient µa of tissue (which is related to theabsorption probability per unit distance) at any wavelengthλ is due to the superposition of the contributions from itsconstituents, where each contribution is given by the prod-uct of the specific absorption ε and the concentration c ofthe constituent. In an equation:

450 IV Breast Carcinoma

Ch4.16-P370468.qxd 3/15/07 5:51 PM Page 450

( ) ca i ii=n m f m/ ^ h (1)

In a simplified description of breast tissue, we can con-sider four main constituents absorbing significantly in thered and NIR, namely, oxy-hemoglobin (HbO2), deoxy-hemoglobin (Hb), water, and lipids. Their specific absorp-tion spectra are known from the literature. Thus, bymeasuring the absorption of breast tissue at a minimum offour different wavelengths, one can estimate the concentra-tion of each constituent.

Scattering is essentially due to the presence of interfaces,or refractive index discontinuities, at a microscopic level.For visible and NIR light, scattering is believed to originatemostly from cell nuclei and subcellular organelles. The scat-tering properties are affected by both the size and density ofthese scattering centers. So their assessment can provideinformation on the microscopic structure of tissue and, inparticular, on the local density of cellular nuclei andorganelles. The wavelength dependence of the transportscattering coefficient µs

′ can be expressed as follows using asimple empirical approximation to Mie theory:

s ( ) a b=n m m-l (2)

This description is in agreement with the experimentalfinding that the transport-scattering coefficient decreasesprogressively upon increasing wavelength, with no charac-teristic peaks. The scattering amplitude a provides informa-tion on the density of the scattering centers (higher values ofa correspond to denser tissues), while the scattering power bis related to their size (smaller scattering centers lead tosteeper slopes).

Because both absorption and scattering contribute tolight attenuation, a direct visualization of the transmittedlight as performed in diaphanography cannot enable the dis-crimination between the two optical phenomena, which haveindependent origins and can even compensate each other.For example, a highly vascularized region of low-density tis-sue combines a higher absorption associated with a higherblood concentration and a lower scattering associated with alower density. These two effects tend to balance each other,leading to overall low contrast, if any. Hence, the possibilityof discriminating between absorption and scattering is ben-eficial and, at least in principle, could increase the sensitiv-ity of optical techniques. Providing more information on thenature of the detected abnormality could also aid its identi-fication, thus affecting positively even the specificity tocancer detection. The assessment of the optical properties(i.e., absorption and scattering coefficients) of tissue in vivoin a clinical environment has become technically feasible inthe last two decades. This implies the potential to derivenoninvasively information on tissue composition and struc-ture that can be profitably used for diagnostic purposes.Consequently, the biomedical community has shown arenewed interest in optical mammography, that is, optical

imaging for the detection and characterization of breastlesions.

Principles of Optical Mammography

Breast imaging approaches can be classified into thosebased on a direct projection of optical data and those basedon solution of the inverse imaging problem of tomography.The latter approaches yield a more rigorous spatial recon-struction of the breast optical properties, but are morecomplex in terms of data acquisition, analysis, and interpre-tation. Consequently, instruments for breast optical tomog-raphy have been developed only recently, and only initialstudies on human subjects have been reported. By contrast,larger human studies based on projection imaging have beenperformed over the last 20 years.

For projection imaging, the breast is typically positionedbetween plane parallel plates, similar to what is done in con-ventional X-ray mammography, but with a much milderdegree of compression so as not to cause discomfort to thepatient even when the full examination requires severalminutes to be completed. As already in the case ofdiaphanography, the light illuminates one side of the breastand the transmitted light is collected on the opposite side.However, discrete wavelengths are typically used, not broadband light. This is done to take advantage of the differentabsorption properties of tissue constituents at differentwavelengths and derive their concentrations, as describedearlier. Moreover, the breast is not fully illuminated. Thelight is generally coupled to an optical fiber that provides arelatively small (1 to 2 mm) illumination spot on the breast,and the light transmitted through the breast is collected withanother optical fiber on the opposite side of the breast.Images are built by raster-scanning the two fibers in tandemover the compressed breast and collecting data every 1 to 2mm. This measurement step determines the pixel size of theimages. This may seem to set an undesired technical limit tothe spatial resolution of optical images but this is not thecase. In fact, at red and NIR wavelengths, the attenuation inbreast tissue is dominated by scattering. Contrary to whatoccurs with X-rays, which mostly propagate straight throughtissue, optical photons undergo hundreds of scatteringevents per centimeter traveled within tissue. So, a collimatedlight beam injected into tissue rapidly broadens upon prop-agation. As a consequence, the shadow cast by an opticalinhomogeneity (e.g., a region of altered vascularization)is expected to appear bigger than the real size of the inho-mogeneity, and the effect is more marked for deeper inho-mogeneities. This sets a physical limitation to the spatialresolution that can be achieved by imaging at optical wave-lengths.

The spatial resolution of optical mammography cannotbe easily quantified because it depends on several factors

16 Optical Mammography 451

Ch4.16-P370468.qxd 3/15/07 5:51 PM Page 451

(optical properties of the inhomogeneity and surroundingtissue, optical contrast, depth of the inhomogeneity).However, as a rule of thumb, we could say that an accurateestimate of the size is possible typically around 1 cm andabove. Smaller objects, down to a few millimeters, can stillbe detected, provided that their optical contrast is largeenough, but their image will be significantly bigger thantheir real size, thus preventing any accurate estimate of theirdimensions. Consequently, optical imaging cannot competewith X-ray mammography on the ground of morphologicinformation. In particular, it will not be possible to visualizesmall calcifications that are a key element for diagnosingmalignant lesions in X-ray images. However, optical datacan provide different pieces of information. Specifically,functional information is available through the assessmentof total hemoglobin content and oxygen saturation, andpotentially of other constituents and related roles.

For breast optical tomography, the geometry of illumi-nation and collection is more complex than for projectionimaging, with a number of possible arrangements. For aparallel-plate geometry (similar to the case of projectionimaging), more illumination and collection points are usedat the planes of illumination and collection, whereas for acircular geometry, arrays of illumination and collection opti-cal fibers are arranged around the pendulous breast. Thelarger number of illumination and collection points yieldsdata that is suitable for tomographic image reconstruction ofthe breast optical properties.

Optical measurements are noninvasive, and safe, and donot cause any discomfort to the patient. The optical instru-mentation is relatively simple and cost-effective, as com-pared to other diagnostic imaging equipment that isroutinely in a clinical setting. Consequently, optical imagingcould be effectively applied even as a complementarytechnique. Moreover, promising preliminary results haverecently been achieved in monitoring neoadjuvantchemotherapy, where repeated measurements can be per-formed with no risk for the patient, thus potentially allowingthe optimized development of individual therapeutic proto-cols.

Continuous-wave Approaches: DynamicMeasurements and Spectral Information

While continuous-wave approaches to optical mammog-raphy are conceptually similar to the diaphanography tech-niques of the 1980s, a number of technical advances in thedata collection and data analysis have been introduced sincethe 1990s to generate much richer functional informationwith respect to the transillumination images of diaphanogra-phy. The term continuous-wave indicates that the lightsource emission is constant with time. On the one hand, thisfact limits the information content of the measured data to

the overall attenuation (or optical density) contributed by thecombination of absorption and scattering events withinbreast tissue. On the other hand, continuous-wave methodsare technologically straightforward, provide a high signal-to-noise ratio, and ideally lend themselves to real-timedynamic measurements and to spectral measurements over abroad, continuous spectrum.

The real-time measurement of dynamic processes withinthe breast has the potential to provide novel functional infor-mation that was not previously sensed by other diagnosticimaging modalities. For example, the oscillatory hemody-namics associated with arterial pulsation or respiration mayreflect the local vascular impedance and compliance, whichcan in turn be affected by cancerous modifications.Furthermore, the dynamic features of the return to equilib-rium in response to a mechanical perturbation (for example,a transient application of pressure to the breast) may revealspatial patterns that can be associated with the presence ofcancerous lesions.

We have already discussed how multiwavelength data canprovide functional and metabolic information, which repre-sents the greatest promise of optical mammography fordiagnostic imaging and distinguishes it from X-ray mam-mography and ultrasonography of the breast. The measure-ment of a continuous optical spectrum, as readily affordedby continuous-wave methods, is a most effective way toidentify the relative concentrations of the various absorbingspecies in breast tissue. Within the spectral region consid-ered in optical mammography, which is typically within the600–1000 nm wavelength range, the absorption of deoxy-hemoglobin decreases with wavelength, with the exceptionof a peak at ~758 nm; the absorption of oxy-hemoglobinshows a broad valley with a minimum at ~692 nm and abroad peak with a maximum at ~924 nm; the absorption ofwater shows a relatively strong peak at ~975 nm; and theabsorption of lipids shows a peak at ~924 nm. The scatter-ing spectrum of breast tissue is featureless and decreaseswith wavelength (see Eq. (2)). It has a wavelength powerdependence that is typically in the range λ−0.4−λ−1.5 (Shahet al., 2004), which is a weaker wavelength dependence thanthe Rayleigh limit (~λ−4) for particles that are much smallerthan the wavelength. Even though single-wavelength, con-tinuous-wave measurements are generally not able to dis-criminate absorption from scattering contributions, fullspectral data can accomplish such a discrimination by takingadvantage of the featureless scattering spectrum of breasttissue. It is well-established that cancer is associated with ahigher concentration of hemoglobin in breast tissue (Fantiniet al., 1998; Tromberg et al., 2000; Grosenick et al., 2003;Dehghani et al., 2003), while it is still unclear whetherhemoglobin saturation provides a reliable intrinsic source ofcontrast for cancer. With regard to water and lipids, casestudies have indicated that cancer, relative to healthy breasttissue, typically has a higher water concentration (Tromberg

452 IV Breast Carcinoma

Ch4.16-P370468.qxd 3/15/07 5:51 PM Page 452

et al., 2000; Jakubowski et al., 2004) and a lower lipids con-tent (Jakubowski et al., 2004).

Time-resolved Approaches

Continuous-wave measurements of the attenuation oflight transmitted through the breast do not allow one to fullyexploit the diagnostic potential of optical mammographythat is associated with separate measurements of absorptionand scattering properties of breast tissue. Time-dependentmethods, where the light source emission is not constantwith time and the optical detection is time-resolved, affordthe measurement of absorption and scattering features ofbreast tissue. Time-resolved approaches are implemented inthe time domain or in the frequency domain. These twoimplementations differ in the instrumentation used, but thedata collected in the time domain and frequency domainare mathematically related by a temporal Fourier transfor-mation.

In time-domain measurements, a short light pulse (~100ps duration) is injected into the tissue. Scattering andabsorption events occurring during propagation through tis-sue cause attenuation, delay, and broadening of the injectedpulse. From a qualitative point of view, we can say that thescattering essentially delays the detected pulse, as each scat-tering event changes the direction of photon propagation.Thus, photons move along “zigzag” trajectories that aremuch longer than the distance between the injection and thedetection points. So photon detection is delayed: the higherthe scattering, the longer the delay. The absorption deter-mines how steep the temporal tail of the detected pulse is.We see the effects of the absorption mostly at long times, onthe pulse tail, because the longer the photons stay in themedium, the higher the probability they undergo an absorp-tion event. Thus, strong absorption means that many photonsare removed from the temporal tail of the pulse and its slopebecomes steeper. This holds qualitatively, but to get a quan-titative estimate of the absorption and scattering properties,we need a suitable theoretical model of light propagationthat correlates the shape and delay of the transmitted pulseto the absorption and scattering properties of breast tissue.Generally, the diffusion approximation to the radiativetransport theory (Patterson et al., 1989) is used, as it pro-vides a simple analytical solution that can be readily appliedfor the interpretation of clinical data. In its simplest deriva-tion, the model holds only for a homogeneous medium. Soit does not allow the estimate of local values of the opticalproperties. It can only provide average values measured overthe light path between the injection and the collection point.More realistic theories that describe the heterogeneity ofbreast tissue are just starting to be developed, and modelsthat take into account at least the presence of a localizedinhomogeneity (i.e., a pathologic lesion) have been applied

only recently, and not routinely yet, for the interpretation ofpatient data (Torricelli et al., 2003; Grosenick et al., 2005).

The idea of optical diagnosis of breast lesions relies onoptical contrast, on the different optical properties of lesionand surrounding tissue. This clearly violates the hypothesisof homogeneous medium. Moreover, the healthy breast tis-sue itself is markedly heterogeneous. Even under such con-ditions, the diffusion approximation still provides effectiveinformation on the scattering properties. On the contrary,absorption images are of difficult use because the inade-quacy of the model limits the contrast and spatial resolution,thus hindering the detection of lesions, especially smallones. However, the availability of the time distribution of thetransmitted photons can have direct applications, as the scat-tering and absorption events modify the pulse shape in a dif-ferent way and at different times. Consequently, aconvenient temporal selection of the detected photons canyield information on the optical properties. In particular,early arriving photons, on the raising edge of the pulse, aremostly (even though not only) affected by the scatteringproperties. Conversely, late-arriving photons, on the tail ofthe detected pulse, are mostly sensitive to the absorptionproperties. Thus, if only “late” photons are selected from thetransmitted pulse at each measurement position during thebreast scan, special intensity images—so called delayedgated images—can be built. The intensity in each “pixel” isrelated to the absorption properties in the correspondingposition: high transmitted intensity indicates low absorp-tion. The method is clearly not quantitative, as the absorp-tion coefficient is not assessed. However, it is relativelysimple, not requiring data analysis based on a theoreticalmodel, and allows one to compare two locations and deter-mine where the absorption is higher or lower and whetherthe difference is small or large. Such a procedure is com-monly used to detect spatial changes in the absorption prop-erties and to identify areas of abnormal absorption, whereasdiffusion theory is used to translate time-domain data intoscattering images.

Time-domain optical mammography has been applied inclinical trials (Grosenick et al., 2005; Taroni et al., 2005)according to the measurement scheme where the slightlycompressed breast is raster-scanned in a transmission geom-etry, and time-domain data are collected at every measure-ment position. Spectral information is obtained by injectingpicosecond pulses at different wavelengths and collectingindependently each of the transmitted pulses. As alreadydescribed, time-gated intensity images are routinely used totrack absorption changes as a function of position. If theimaging wavelengths are chosen so that each wavelengthisolates the contribution of a specific constituent, namely, asingle constituent absorbs at each wavelength, then eachimage can be used to investigate the spatial distribution ofa different constituent. In practice, it is usually difficult tofind wavelengths where just one constituent absorbs. So, it

16 Optical Mammography 453

Ch4.16-P370468.qxd 3/15/07 5:51 PM Page 453

is not possible to isolate single constituents, but still, withproper choice of the imaging wavelengths, each of them candominate a different image. In particular, Politecnico diMilano (Milan, Italy) has developed a prototype that oper-ates at 4 to 7 wavelengths between 637 and 985 nm (Taroniet al., 2005). Using that instrument, it was shown thatimages at wavelengths shorter than 685 nm are dominatedby deoxy-hemoglobin, at 916 nm by lipids, and at 975 nmby water, while 785 nm enhances the sensitivity to oxy-hemoglobin. An example is shown in Figure 116, which dis-plays images acquired at 683, 785, 916, and 975 nm froma 55-year-old patient with a 1.8 cm invasive ductal carci-noma in her left breast. The cancer is detected because of itsmarked vascularization that causes strong absorption at thetwo shorter wavelengths (Figs. 116a and 116b). Moreover,the uniform and dark appearance of the image at 916 nm(Fig. 116c) indicates a high lipid content, revealing the adi-pose nature of the breast, in agreement with what derivedfrom the X-ray mammogram (Fig. 116e). Further details onimage interpretation are reported in Section F.

The frequency-domain approach is based on modulat-ing the intensity of the light source (at a frequency f that istypically in the order of 100 MHz) and performing phase-sensitive detection of the modulated optical signal. Onecan fully describe the modulated optical signal using threeparameters, namely, the average intensity (DC intensity),and the amplitude (AC amplitude) and phase (Φ) of theintensity oscillations. These three parameters can providesufficient information to characterize both the absorptionand scattering properties of tissue. In particular, the phasemeasurement is directly associated with the time-delayexperienced by the probing intensity-wave in tissue, which

is also directly measured in the time-domain approach.Frequency-domain instruments have been developed fordirect projection imaging of the breast (Fantini et al.,2005), as well as optical tomographic imaging of the breast(Dehghani et al., 2003). A frequency-domain prototypeoriginally developed by Siemens AG, Erlangen, Germany(Götz et al., 1998), has produced a clinical data set of opti-cal mammograms in a planar transmission geometry.Figure 117 shows a picture of this prototype (panel (a)) andtwo representative optical mammograms (panels (b) and(c)) obtained on a 53-year-old patient with a 3 cm invasiveductal carcinoma in her left breast. The optical mammo-grams shown in Figure 117 (panels (b) and (c)) are directprojection images of the slightly compressed left breasttaken in a craniocaudal (cc) projection. The total timerequired to scan the breast is 2 to 3 min. The frequency-domain optical data have been initially processed with analgorithm designed to enhance tumor contrast by minimiz-ing the effects of the breast geometry on the optical data.Then a spatial second derivative algorithm is applied toenhance the spatial information content of the image andthe visualization of blood vessels (Fig. 117b). Finally, dataat four wavelengths (690, 750, 788, and 856 nm) are com-bined to yield a measure of tissue saturation (StO2) thatresults in an oxygenation image (Fig. 117c) The opticalmammograms in Figure 117 illustrate the potential of thisimaging technique to detect angiogenic signatures andoxygenation data associated with breast tumors. In partic-ular, the invasive ductal carcinoma of Figure 117 (indi-cated by the arrow in panels (b) and (c)) is associated witha high density of blood vessels (Fig. 117b) and low levelsof oxygenation (Fig. 117c).

454 IV Breast Carcinoma

(a) 683 nm (b) 785 nm (c) 916 nm (d) 975 nm (e) X-ray image

Figure 116 Late-gated intensity images at 683 nm (a), 785 nm (b), 916 nm (c) and 975 nm (d), and X-raymammogram (e) of the left breast (oblique view) of a 55-year-old patient bearing an invasive ductal carcinoma(max. diameter = 1.8 cm), indicated by the red arrow. The images were acquired with the time-resolved multiwavelength optical mammograph developed by Politecnico di Milano (Milan, Italy).

[AU1]

Ch4.16-P370468.qxd 3/15/07 5:51 PM Page 454

Interpretation of Optical Mammograms

A number of structures of the healthy breast can be iden-tified in optical images. First, blood vessels and highly vas-cularized regions, like those surrounding the lactiferousducts in the nipple area, are detected due to the strong hemo-globin absorption at short wavelengths (below 800 nm). Themammary gland is characterized by marked absorption at975 nm, possibly due to its high water content as comparedto the surrounding more adipose breast tissue, even thougha contribution could also come from collagen, which showssignificant absorption at 975 nm. Moreover, lipids have acharacteristic absorption peak around 924 nm. Thus, opticalimages collected around this wavelength of 924 nm provideinformation on lipid distribution, and regions that appeardark in those images, due to strong absorption, show goodcorrespondence with areas that are relatively transparent toX-rays. The scattering images of healthy breasts are gener-ally uniform, except for the mammary gland that in somecases reveals slightly lower scattering, especially at thelongest wavelengths.

Concerning breast lesions, cancers are usually identifiedthrough the detection of associated neovascularization.Thus, they appear as strongly absorbing areas at short

wavelengths. The contrast is often higher at 637–685 nmthan at 780–785 nm, suggesting the presence of deoxy-genated blood. However, low oxygenation has not yet beenproven to be a reliable index for identifying malignantlesions. These qualitative observations are based on thevisual inspection of images acquired at different wave-lengths. However, they are confirmed by the quantitativeestimate of blood content and oxygenation in the lesion andsurrounding tissue. Such estimates are obtained applying“inhomogeneous” models of breast tissue that account forthe presence of a localized lesion and allow one to quantifyits optical properties and estimate its composition. Thesemodels indicate that the blood content in the tumor area istypically two to five times higher than in the surroundingtissue. However, such values are often exceeded. On thecontrary, no systematic and reliable findings have yet beenreported for the oxygenation level of breast cancer withrespect to the oxygenation level of healthy tissue or benignbreast lesions.

At least in principle, the scattering images could providediagnostically useful information. Actually, the neoplastictransformation affects the entire tissue architecture, alteringcell density and nuclear volume, degrading the extracellularmatrix upon invasion, and forming a complex network of

16 Optical Mammography 455

low highStO2

(a) (b)

(c)

lcc

lcc

Figure 117 (a) Photograph of a prototype for frequency-domain (70 MHz) optical mammography developed bySiemens AG, Medical Engineering (Erlangen, Germany). The slightly compressed breast is optically scanned to obtain2D projection images at four wavelengths (690, 750, 788, and 856 nm). The scanning time is about 2 min per image.(b) Second-derivative image at 690 nm, and (c) oxygenation image from data at all four wavelengths of the left(l) breast in craniocaudal (cc) view of a 53-year-old patient affected by invasive ductal carcinoma (indicated by thearrow in panels (b) and (c)). Cancer size is 3 cm.

Ch4.16-P370468.qxd 3/15/07 5:52 PM Page 455

new blood vessels. All these processes may modify the scat-tering properties of tissues, especially when poorly differen-tiated invasive lesions develop. Experimentally, changes inscattering are often observed, especially when the lesioninvolves a significant volume. The detection and identifica-tion of cysts are performed relying on their liquid nature thatleads to low scattering. Cysts may be of several types, con-taining a clear fluid, a turbid liquid, or even big floating par-ticles. The effect of these various structures on the slope ofthe scattering spectrum is different. Consequently, thehigher or lower contrast at different wavelengths providesinformation on the nature of the cyst. For example, a markedincrease in contrast at long wavelengths, corresponding to avery steep scattering spectrum, will suggest the presence ofa clear fluid. In some cases, specific absorption featuresmay also identify fluid-filled cysts, as a result of their rela-tively high water concentration, low lipid concentration, andhigh concentration of deoxygenated blood.

The detection of fibroadenomas is often more challeng-ing for optical mammography. When identified, fibroadeno-mas are generally characterized by high absorption around975 nm and sometimes even in the red spectral region(637–685 nm). The marked absorption at 975 nm is likelyrelated to their high water content, but it might also be due,at least in part, to collagen. In the early stages of a clinicaltest of the only time-domain instrument currently featuringwavelengths longer than 900 nm (Taroni et al., 2005), dataacquisition was characterized by limited signal levels atwavelengths longer than 900 nm. Such technical limitationhas likely reduced the potential for detecting fibroadenomasin that clinical test. However, instrumental upgrade is ongo-ing and is expected to improve the diagnostic potential forfibroadenomas in the near future.

Optical images are routinely acquired in two views:craniocaudal and either mediolateral (90˚) or oblique (45˚).If localization of the lesion is required in both views, thedetection rate (sensitivity) of optical mammography isaround 80% for both cancers and cysts. As alreadyobserved, the detection of fibroadenomas is often difficult,with only 39% of fibroadenomas identified in both views.If the criterion for positive classification is relaxed to lesionlocalization in just one view, the sensitivity increases sig-nificantly, reaching 92 to 96% for cancers and 90% forcysts.

On average, the optical contrast of detected cancersincreases progressively with their size, but no such correla-tion with size has been reported for other lesion types. Thisdifferent trend observed for malignant and benign lesions islikely due to the fact that cancers are detected in intensityimages at short wavelengths, where all blood-reach struc-tures, not only the neovascularization associated with tumordevelopment, are highlighted and can hamper the detectionof the lesion. Thus, a bigger lesion size can significantlyincrease the optical contrast. On the contrary, cysts are

detected in scattering images, which are rather uniform forhealthy breasts. Thus, no main dependence on the lesion sizeand generally higher contrast can be expected. Demographicparameters, such as age and body mass index (BMI), seemto have no influence on lesion detection or on the contrastfor detected lesions, either malignant or benign ones ofany kind.

Prospects of Optical Mammography

X-ray mammography, the current gold standard for breastcancer screening, is less effective in women younger than50, who have radiographically dense breasts, leading to highrates of both false-negative and false-positive cases. By con-trast, no clear dependence on age was observed for opticalimaging in terms of cancer detection rate or image contrast.The possible effect of breast density was also investigated.Five mammographic parenchymal patterns were identifiedfollowing Tabàr’s classification (Gram et al., 1997). Densebreasts (type IV and V) were compared with adipose breasts,more transparent to X-rays (type II and III). The detectionrate is a few percent points higher for dense breasts, withslightly lower average value of the detection contrast.However, the difference between the two categories is notsignificant (Taroni et al., 2005). Because most findings arebased on retrospective studies, involving patients withlesions previously identified in X-ray mammograms, thereis no information on how optical mammography performs inpatients with false-negative X-ray results. Consequently, theresults might be somehow biased. Nevertheless, there arestrong indications that optical imaging is not negativelyaffected by the radiological density of breast. For this rea-son, optical mammography may find a niche of clinicalapplicability in a population of younger women, where X-ray mammography is not applicable or suffers fromdegraded performance.

It is also envisioned that optical mammography mayeffectively complement X-ray mammography. First, theinformation about hemodynamic, oxygenation, andwater/lipids composition provided by optical mammographyis complementary to the fine structural information pro-vided by X-ray mammography. The combination of thesecomplementary pieces of diagnostic information has thepotential to result in a more effective breast imaging modal-ity than X-ray mammography alone. Second, the opticalscattering spectrum correlates with breast density.Consequently, a prescreening optical mammogram mayidentify breasts at high risk for cancer and indicate cases inwhich X-ray exposure should be avoided if conventionalmammography is not expected to be effective.

There is also promise in the combination of ultrasoundand optical techniques. On the one hand, it is possible tocomplement the information content of optical mammography

456 IV Breast Carcinoma

Ch4.16-P370468.qxd 3/15/07 5:52 PM Page 456

16 Optical Mammography 457

and ultrasound imaging, similarly to the way that it is envi-sioned to complement the information content of optical andX-ray mammography. On the other hand, ultrasound andlight may be combined into a truly hybrid diagnostic tooleither by using a focused ultrasound beam to label or tagoptical photons that have traveled through the ultrasoundfocal volume (ultrasonic tagging of light), or by using pho-toacoustics to generate high-frequency pressure waves(ultrasound) as a result of localized areas of increased opti-cal absorbance. It has also been suggested that the rich spa-tial information provided by magnetic resonance imaging(MRI) can provide crucial a priori information for theimplementation of optical tomographic reconstruction algo-rithms. In this sense, MRI and optical mammography canalso be combined into a hybrid imaging tool.

Another potential clinical role of optical mammographyis in the area of monitoring the effectiveness of therapeuticprocedures and performing post-treatment follow-up. Thispotential results from features of optical mammographysuch as its safety, lack of discomfort, noninvasiveness, real-time capability, implementation in portable instrumentalunits, and cost-effectiveness.

While there is an obvious emphasis on basing opticalmammography on the intrinsic optical contrast provided bythe human breast (mostly from oxy-hemoglobin, deoxy-hemoglobin, water, and lipids), research efforts are alsobeing aimed at assessing the potential offered by extrinsicoptical contrast agents. Although extrinsic contrast agentsintroduce the need for an intravenous injection, thus makingthe procedure invasive, they can offer unprecedented oppor-tunities and possibly lead to a more powerful approach (interms of both sensitivity and specificity) to the opticaldetection of breast cancer. For example, it is possible todetect specific enzyme activity by using auto-quenchednear-infrared fluorescence probes, whose fluorescenceemission is restored in the presence of enzymes that areoverexpressed by tumors. Other approaches use near-infrared fluorescent dyes (typically, the clinically approvedindocyanine green or structurally related dyes) that exhibitpreferential accumulation in cancerous tissue. Some animalstudies have shown promising results of cancer detectionbased on extrinsic contrast agents, but more research isneeded to fully appreciate the potential of this approach inhumans.

In conclusion, optical mammography is a potentiallypowerful imaging modality for the human breast, whichprovides diagnostic information that is not available fromother current imaging tools. Optical mammography mayplay an important clinical role as a stand-alone techniquefor breast cancer detection (especially in younger women)or for follow-up to treatment, and can effectively comple-ment other diagnostic imaging modalities such as X-raymammography, ultrasound imaging, and magnetic reso-nance imaging. The potential of combining multiple imag-

ing tools for the detection and diagnosis of cancer isenormous, and optical methods can play an important rolein developing such synergistic combinations of imagingtools.

Acknowledgments

We acknowledge support from the National Institutes of Health(Grant CA95885) and the National Science Foundation (Award BES-93840).

References

Bartrum, R.J., and Crow, H.C. 1984. Transillumination lightscanning todiagnose breast cancer: a feasibility study. AJR 142:409–414.

Fantini, S., Walker, S.A., Franceschini, M.A., Kaschke, M., Schlag, P.M.,and Moesta, K.T. 1998. Assessment of the size, position, and opticalproperties of breast tumors in vivo by non-invasive optical methods.Appl. Opt. 37:1982–1989.

Fantini, S., Heffer, E.L., Pera, V.E., Sassaroli, A., and Liu, N. 2005. Spatialand spectral information in optical mammography. Technol. CancerRes. Treat. 4:471–482.

Dehghani, H., Pogue, B.W., Poplack, S.P., and Paulsen, K.D. 2003.Multiwavelength three-dimensional near-infrared tomography of thebreast: initial simulation, phantom, and clinical results. Appl. Opt.42:135–145.

Götz, L., Heywang-Köbrunner, S.H., Schütz, O., and Siebold, H. 1998.Optische mammographie an präoperativen patientinnen. Akt. Radiol.8:31–33.

Gram, I.T., Funkhouser, E., and Tabar, L. 1997. The Tabàr classification ofmammographic parenchymal patterns. Eur. J. Radiol. 24:131–136.

Grosenick, D., Moesta, K.T., Wabnitz, H., Mücke, J., Stroszczynski, C.,Macdonald, R., Schlag, P.M., and Rinneberg, H. 2003. Time-domainoptical mammography: initial clinical results on detection and charac-terization of breast tumors. Appl. Opt. 42:3170–3186.

Grosenick, D., Wabnitz, H., Moesta, K.T., Mucke, J, Schlag, P.M., andRinneberg, H. 2005. Time-domain optical mammography (part II):optical properties and tissue parameters of 87 carcinomas. Phys. Med.Biol. 50:2451–2468.

Jakubowski, D.B., Cerussi, A.E., Bevilacqua, F., Shah, N., Hsiang, D.,Butler, J., and Tromberg, B.J. 2004. Monitoring neoadjuvantchemotherapy in breast cancer using quantitative diffuse optical spec-troscopy: a case study. J. Biomed. Opt. 9:230–238.

Monsees, B., Destouet, J.M., and Totty, W.G. 1987. Light scanning versusmammography in breast cancer detection. Radiology 163:463–465.

Ohlsson, B., Gundersen, J., and Nilsson, D.M. 1984. Diaphanography: amethod for evaluation of the female breast. World J. Surg. 4:701–707.

Patterson, M.S., Chance, B., and Wilson, B.C. 1989. Time-resolvedreflectance and transmittance for the noninvasive measurement of tissueoptical properties. Appl. Opt. 28:2331–2336.

Shah, N., Cerussi, A.E., Jakubowski, D., Hsiang, D., Butler, J., andTromberg, B.J. 2004. Spatial variations in optical and physiologicalproperties of healthy breast tissue. J. Biomed. Opt. 9:534–540.

Taroni, P., Torricelli, A., Spinelli, L., Pifferi, A., Arpaia, F., Danesini, G, andCubeddu, R. 2005. Time-resolved optical mammography between 637and 985 nm: clinical study on the detection and identification of breastlesions. Phys. Med. Biol. 50:2469–2488.

Torricelli, A., Spinelli, L., Pifferi, A., Taroni, P., and Cubeddu, R. 2003. Useof a nonlinear perturbation approach for in vivo breast lesion character-ization by multi-wavelength time-resolved optical mammography. Opt.Expr. 11:853–867.

Ch4.16-P370468.qxd 3/15/07 5:52 PM Page 457

Tromberg, B.J., Shah, N., Lanning, R., Cerussi, A., Espinoza, J., Pham, T.,Svaasand, L., and Butler, J. 2000. Non-invasive in vivo characterizationof breast tumors using photon migration spectroscopy. Neoplasia2:26–40.

Wallberg, H., Alveryd, A., Bergvall, U., Nasiell, K., Sundelin, P., and Troell,S. 1985. Diaphanography in breast carcinoma: correlation with clinicalexamination, mammography, cytology and histology. Acta Radiol.Diagn. 26:33–44.

458 IV Breast Carcinoma

AU1: What is Sention F?

Ch4.16-P370468.qxd 3/15/07 5:52 PM Page 458