is

Ma

b

c

a

A

R

R

1

A

K

G

A

M

F

1

OeratuipncnoC(

0h

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 715–723

jo ur n al hom ep age : www.int l .e lsev ierhea l th .com/ journa ls /cmpb

GAIT: An interactive accelerometer based gait analysisystem

ingjing Yanga, Huiru Zhenga,∗, Haiying Wanga, Sally McCleanb, Dave Newell c

School of Computing and Mathematics, University of Ulster, N. Ireland, UKSchool of Computing and Information Engineering, University of Ulster, N. Ireland, UKAnglo-European College of Chiropractic, 13-15 Parkwood Road, Bournemouth, Dorset BH5 2DF, UK

r t i c l e i n f o

rticle history:

eceived 19 October 2010

eceived in revised form

9 November 2011

ccepted 11 April 2012

eywords:

ait analysis

a b s t r a c t

This paper presents a software program (iGAIT) developed in MATLAB, for the analysis of

gait patterns extracted from accelerometer recordings. iGAIT provides a user-friendly graph-

ical interface to display and analyse gait acceleration data recorded by an accelerometer

attached to the lower back of subjects. The core function of iGAIT is gait feature extraction,

which can be used to derive 31 features from acceleration data, including 6 spatio-temporal

features, 7 regularity and symmetry features, and 18 spectral features. Features extracted

are summarised and displayed on screen, as well as an option to be stored in text files for

further review or analysis if required. Another unique feature of iGAIT is that it provides

ccelerometer

ATLAB

eature extraction

interactive functionality allowing users to manually adjust the analysis process according

to their requirements. The system has been tested under Window XP, Vista and Window 7

using three different types of accelerometer data. It is designed for analysis of accelerometer

data recorded with sample frequencies ranging from 5 Hz to 200 Hz.

discriminate between nine subjects with medial osteoarthrtic

. Introduction

ver the past decade, there have been considerable researchfforts to employ accelerometers in healthcare. Some of theesulting applications use accelerometers to monitor dailyctivities and classify movements between patients and con-rols [1,2]. More recently, accelerometers have been widelysed in gait pattern analysis [3–7] for different purposes

ncluding monitoring physical function of the elderly [2,8,9],rediction of fall risk [10,11], evaluation of the progress ofeurodegenerative disease [12,13], and assessment of the out-ome of therapy [14]. Accelerometer based gait data provide aovel way to measure gait patterns. Compared with traditional

ptical gait analysis systems, such as CODA (Codamotion,harnwood Dunamics Ltd., Rothley, England) and Vicon MX

Vicon, Oxford, England), accelerometer-based approaches

∗ Corresponding author. Tel.: +44 02890366591; fax: +44 02890366216.E-mail address: h.zheng@ulster.ac.uk (H. Zheng).

169-2607/$ – see front matter © 2012 Elsevier Ireland Ltd. All rights resttp://dx.doi.org/10.1016/j.cmpb.2012.04.004

© 2012 Elsevier Ireland Ltd. All rights reserved.

exhibit several advantages. For example, they are lower cost,portable, small in size, easy to operate and as a consequence,can be employed in real life environments to support rehabil-itation [15] and physical activity monitoring [1].

Many types of accelerometers have been reported as beingused in monitoring gait [1]. The simplest application ofaccelerometers is as a step counter; however it has been foundthat other gait parameters, such as cadence, speed, or symme-try, can be extracted from accelerometer data [16–20]. Turcotet al. [21] used two triaxial accelerometers (ADXL329) com-bined with gyroscopes and reflective marks to estimate thetibial and femoral linear accelerations. In total eight statisticalparameters extracted from the acceleration data were used to

knees and nine asymptomatic control subjects. Zhou et al. [14]used a system composed of five biaxial IDEEA accelerometers(MiniSun, CA, USA) mounted on the chest, thighs and feet to

erved.

716 c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 715–723

Fig. 1 – The graphical user interface of iGAIT.

measure gait patterns of subjects following total hip arthro-plasty. Several features related to speed, stride length, singlelimb support, cycle period and swing power were extracted asgait pattern indices. In the research conducted by Tura et al.[22], a MTx accelerometer (XSENS Technologies B.V.) was wornon the subjects thorax to assess gait symmetry and regularityof transfemoral amputees. DynaPort MiniMod (McRobert BV,The Hague, The Netherlands) has been employed to study thegait pattern in ageing people [9], youths [3], and children [7,23].These accelerometer systems have their respective character-istics, for example different data formats, sampling frequency,number of accelerometers and placement position [1]. In thisstudy we have used a single triaxial accelerometer attached tothe lower back. The position is an approximation of the centreof mass (COM) of the body during walking and upright stances[24]. This placement is convenient and less intrusive for thesubject compared with other positions, such as chest, neck,thigh and head. In addition, it has been shown that many keygait parameters, such as cadence, step length, speed, symme-try and regularity, can be obtained from such accelerometerdata [3–7].

To the best of our knowledge, there is no publicly avail-able software for the analysis of lower back accelerationdata during walking. Most researchers use in-house software

to analyse gait patterns from the acceleration data. Someaccelerometer manufacturers provide gait data analysis soft-ware packages, which are designed to process informationextracted from their own products. For example, McRobert BV

provides a gait analysis service, but only to their registeredcustomers who pay an annual fee. After users upload acceler-ation data, the service extracts the gait parameters, save theresults in a PDF file and distributes it to the users [25]. Althoughthis type of gait data analysis service is convenient it remainsa ‘black box’ to the user, potentially limiting confidence in theanalysis and the application of user specific needs, particu-larly when data are collected from subjects with abnormal gaitpatterns.

In this paper, we present a computer aided gait assess-ment tool, iGAIT, for the analysis of gait data collected byan accelerometer. iGAIT is implemented in MATLAB and canbe run in Windows XP, Vista and Windows 7. It provides aninteractive and user friendly platform to visualise accelerationdata. A total of four types of gait features: spatio-temporal, fre-quency domain, regularity and symmetry can be derived. Themain advantage of iGAIT includes its capacity for analysingaccelerometer data collected by accelerometer devices fromdifferent manufacturers, such as Dynaport MiniMod, smartmobile phones (such as HTC Touch Pro) with accelerometersand MTx (Xsens Technolgies B.V.). Another unique feature isthat it allows users to select specific gait cycles for furtherexamination. The results (gait features extracted) can be savedas a text file that can be directly subjected to further analy-

sis (statistical or machine learning). iGAIT is designed for theanalysis of accelerometer data recorded with sample frequen-cies ranging from 5 Hz to 200 Hz, which is the most commonlyused in gait analysis. In addition to the research lab, iGAIT can

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i

Fig. 2 – The overview of the steps taken by iGAIT to analysea

asc[r

tfcf

2

TmagfTc

(Mfira

2

TFte

2

Tep

before and after walking. Therefore, iGAIT allows user selec-

ccelerometer gait data.

lso be used in routine clinical practice. For example, acqui-ition of gait parameters following a short ambulation in alinical setting could be used for the assessment of fall risk26], musculoskeletal pain syndromes [27–30] and monitoringehabilitation intervention [31].

The remainder of this paper is organised as follows: in Sec-ion 2, the interface and functions of the tool are presentedollowed by a case study described in Section 3. The paper isoncluded in Section 4 by a discussion, summary and outlookor future work.

. Program description

he iGAIT software developed within the MATLAB environ-ent is intended to be an accessible tool for the clinician

nd/or researcher who is interested in accelerometer basedait pattern analysis. All functionalities are directly availablerom a graphical user interface (GUI) as illustrated in Fig. 1.he intention is that users are not required to have advancedomputer skills.

The program was entirely developed in MATLAB 2009aThe MathWorks, USA) with the Signal Processing Toolbox and

ATLAB Compiler. All functions are contained in a single m-le. Further, new functions can be added into the program asequired. This structure therefore provides easy extensibilitynd scalability.

.1. Program overview

he procedure for gait analysis includes 6 steps as shown inig. 2, namely, signal loading, parameters setting, the selec-ion of interesting signal regions, gait event detection, featurextraction and results output.

.2. The graphical user interface

he interface of iGAIT is divided into four different pan-ls (Fig. 1): plot panel (top), parameter panel (centre), resultsanel (bottom-left) and demand button panel (bottom-right).

o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 715–723 717

It allows users to load signals, plot signals, set parameters,manually select gait cycles, when necessary, and display gaitfeatures on screen.

2.2.1. Signal loadingThe signal analysed by iGAIT should be collected by a triaxialaccelerometer mounted on the upper body along the verte-bral column, for example the lower back, neck and chest. Theacceleration data are provided as a 3 dimensional numericmatrix saved in a MATLAB (MAT) file, a text file or a MicrosoftExcel sheet. The first column is the acceleration in the verticaldirection (VT), the second in the medial-lateral (ML) directionand the third in the anterior–posterior (AP) direction. The textfile can be saved in a number of formats, including white-space delimited numbers, tab delimited numbers and commadelimited numbers. Hence, iGAIT is flexible enough to anal-yse gait data collected by a range of different accelerometers.After loading, the signals are displayed on the plot panel asshown in Fig. 1.

2.2.2. Parameters inputThere are three parameters that are required to be set bythe user, namely Sample Rate, Distance and Threshold as indi-cated in Fig. 1. In reported studies, accelerometers used inmovement analysis have a wide range of sampling frequency(5–256 Hz) [1,20–22,32]. For example when they are used asstep counters or for monitoring the daily energy expendi-ture, the selection of sampling frequency is not always critical,usually up to 10 Hz [1]. On the other hand, when accelerom-eters are used in gait pattern analysis, they normally havehigher sample rates often above 50 Hz. To facilitate suchanalyses, iGAIT allows the user to set the sampling fre-quency as needed. By default, the sampling frequency is set at10 ms (100 Hz).

In iGAIT, the average walking velocity is calculated by divid-ing walking distance by walking time. Since the distanceinformation cannot be directly derived from the accelerom-eter, this must be inputted by the user.

To detect heel contacts, a threshold with a range of (0,1) needs to be set. This threshold is defined as the ratioto the maximum value of the AP acceleration, for example0.5 indicates the threshold is set at 50% of the maximumAP acceleration. Its default value is set to 0.4 as determinedexperimentally in this paper, where this value allowed correctdetection of all gait events in control subjects. However, whena more irregular pattern is analysed, the threshold should beless than 0.4. The user can test different threshold valuesand find the best one according to the gait event detectionresults.

2.2.3. Selection of signal region of interestAn important feature of iGAIT is that it allows the user toselect regions of interest for a given signal. This function isuseful when the raw data contain redundant segments. Forexample, as shown in Fig. 3, the acceleration data collected bya mobile phone during walking normally include the phases

tion of data segments. This is also essential for some typesof analysis where particular phases of walking attract moreattention. For example, when analysing the gait patterns of

718 c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 715–723

Fig. 3 – The signal before and after selection. (a) The acceleration data collected by accelerometer embbeded in mobilephone, which includes redundant information. (b) The selection of acceleration data without redundant information.

subjects with Parkinson’s disease, the steady walking phaseis of more interest to clinicians and researchers [33,34]. Whenusing iGAIT to analyse gait data of subjects with Parkinson’sdisease, the beginning and the ending part of the walking datacan be discarded so that only the region of interest is selected.

The selected signal is validated before frequency analysis.If the number of samples in the selected acceleration sequenceis less than ten times the sampling frequency, which leads toa frequency resolution of less than 0.1 Hz, a message box willpop up to inform the user.

In an attempt to obtain a clear profile, after the selection ofthe signal region, only the signal from the VT and AP directionsis displayed in the plot panel as shown in Fig. 3(b).

2.2.4. Gait events detectionThe first step of gait analysis is to detect heel contact eventsand to divide the acceleration signal into steps, which aredefined as the interval between the heel contact of foot withthe ground and the heel contact of the other foot. The heelcontacts are detected by peaks preceding the sign changeof AP acceleration [3]. In order to automatically detect aheel contact event, firstly, the AP acceleration is low passfiltered by the 4th order zero lag Butterworth filter whosecut frequency is set to 5 Hz. After that, transitional positionswhere AP acceleration changes from positive to negative can

be identified. Finally the peaks of AP acceleration precedingthe transitional positions, and greater than the product ofa threshold and the maximum value of the AP accelerationare denoted as heel contact events, as shown in Fig. 4. In this

example the default threshold value is set to 0.4; howeveriGAIT allows the user to select different threshold value. Theproposed threshold based heel strike detection algorithm hasbeen tested using the acceleration gait data collected fromabout 100 healthy subjects and 20 subjects with chronic pain.

In some cases, for example the acceleration data collectedfrom subjects with chronic pain, the data are quite irregu-lar [27–30]. In this case, where the program may have beenunable to detect all heel strikes, iGAIT allows the user to addthe omitted heel contact events or remove spurious heel con-tact events by using the mouse to select the related points onthe acceleration plot.

2.2.5. Spatio-temporal featuresAfter the detection of heel contacts, the number of steps isobtained by counting these contacts. Cadence is calculatedusing the number of steps divided by walking time; walkingvelocity is obtained from the walking distance divided bywalking time. Mean step length is obtained from the walk-ing distance divided by the number of steps. The root meansquare (RMS) of acceleration indicates the intensity of motion.The RMS values of the three acceleration directions (VT, APand ML) are calculated respectively using Eq. (1):

RMSd =

√√√√ N∑i=1

(xdi − x̄d)2/N (1)

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 715–723 719

Fig. 4 – The results of gait feature analysis. In the plot panel, the X axis is time; the Y axis is the acceleration. The greentrace is the raw VT acceleration. The red trace is the raw AP acceleration; the blue trace is the AP acceleration after low passfiltering. The blue circles are the transitional positions while the black triangles are the heel contact events. (Fori , the

westaR

2idafdE

P

I

C

nterpretation of the references to color in this figure legend

here xdi (i = 1, 2,. . .,N; d = VT, AP, ML) is the acceleration inither the VT, AP or ML axis, N is the length of the accelerationignal, x̄d is the mean value of acceleration in any axis. Inotal, iGAIT extracts six spatio-temporal features from thecceleration data, namely, cadence, mean step length, velocity,MSVT, RMSAP, and RMSML.

.2.6. Spectral analysisGAIT also extracts six frequency domain features from eachirection of the acceleration data, which measure both themplitude and frequency of body movements. One of theserequency domain features is the integral of the power spectralensity (IPSD), which was estimated by a periodogram usingqs. (2) and (3).

SD(ejω) = 12�N

∣∣∣∣∣N∑

i=1

xie−jωi

∣∣∣∣∣2

(2)

PSD =∫ �

0

PSD(ω)dω (3)

PSD(ω) =∫ ω

0

PSD(ω)dω (4)

reader is referred to the web version of the article.)

where ω (0 ≤ ω ≤ �) is the angular frequency, xi (i = 1, 2,. . ., N) isthe acceleration in either the VT, AP or ML axis, and N is thetotal number of acceleration samples. The frequency with themaximum value of PSD is the second frequency feature calledmain frequency. The other 4 frequency features (Fr1, Fr2, Fr3 andFr4) are the frequencies when PSD is cumulated (CPSD), whichis calculated by Eq. (4), for 50%, 75%, 90% and 99% of IPSD,respectively. In total, iGAIT extracts eighteen frequency fea-tures from the acceleration data, namely, IPSD, main frequency,Fr1, Fr2, Fr3 and Fr4 in each of three directions (VT, ML and AP).

2.2.7. Computation of the regularity and symmetryparametersThe autocorrelation coefficient refers to the correlation of atime series with its own past or future values. iGAIT uses unbi-ased autocorrelation coefficients of acceleration data to scalethe regularity and symmetry of gait [30]. The unbiased esti-mate of autocorrelation coefficients of acceleration data canbe calculated by Eq. (5):

fc(t) = 1N − |t|

N−|t|∑i=1

xixi+t (5)

s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 715–723

-2 0 2-0.5

0

0.5

1

Time Lag (s)

Nor

mal

ised

Aut

ocor

rela

tion

Coe

ffic

ient

s

VT

-2 0 2-0.5

0

0.5

1ML

Time Lag (s)-2 0 2

-0.5

0

0.5

1AP

Time Lag (s)

Fig. 5 – The autocorrelation coefficients of accelerationcurves in three directions, the X axis is time lag (s); the Yaxis is the normalised autocorrelation coefficients; greenstar denotes the step regularity index, red star denotes thestride regularity index. (For interpretation of the referencesto color in this figure legend, the reader is referred to the

720 c o m p u t e r m e t h o d s a n d p r o g r a m

where xi (i = 1, 2,. . ., N) is the acceleration data, fc(t) are auto-correlation coefficients. t is the time lag (t = −N, −N + 1,. . .,0, 1,2,. . ., N). When the time lag t is equal to the periodicity of theacceleration xi, a peak will be found in the fc(t) series.

The autocorrelation coefficients are divided by fc(0) in Eq.(6), so that the autocorrelation coefficient is equal to 1 whent = 0.

NFC(t) = fc(t)fc(0)

(6)

Here NFC(t) is the normalised autocorrelation coefficient,and fc(t) are autocorrelation coefficients. t is the time lag(t = −N, −N + 1, . . ., 0, 1, 2,. . ., N).

The peak value a1 = NFC(t1), which is found at the t1 lag timeto be a step period, indicates the step regularity in a direction.The peak value a2 = NFC(t2), which was found at the t2 lag timeto be a stride period, indicates the stride regularity in a direc-tion. The variance of the stride regularity and step regularityD = |a2 − a1|, can be used as an indicator to measure gait sym-metry. A smaller value of D indicates a more symmetric gait.A larger value of D indicates a more asymmetric gait. In total,iGAIT extracts seven regularity and symmetry features fromthe acceleration data, namely, Step regularity VT, Stride regular-ity VT, Symmetry VT, Step regularity AP, Stride regularity AP, andSymmetry AP and Stride regularity ML.

2.2.8. Results display and savingAll of the four types of features (31 in total) extracted fromthe acceleration data can be saved in a text file, which can beused in further study such as statistical analysis and machinelearning analysis. Eighteen out of thirty one features are alsodisplayed on the screen in the results panel (Fig. 4).

2.3. System specification

The iGAIT system was compiled as a standalone Windowsapplication. iGAIT can run without the Matlab complete instal-lation and requires 400 MB of free hard disk space to installMatlab Compiler Runtime, which is provided with iGAIT. Theprogram needs 120 M memory (RAM) and has been testedunder Windows XP, Vista and Windows 7 with a 2 GHz pro-cessor and display resolution set to 1024 × 768.

3. Sample runs

iGAIT was used to analyse the gait pattern of 15 control sub-jects (6 males and 9 females). A smart mobile phone (HTCTouch HD), in which a tri-axial accelerometer is embedded,was attached to the back of participants by belts. In each trial,participants walked 25 m along a hallway at their preferredwalking speed.

The sampling frequency of the accelerometer was 25 Hz,which is the highest sampling frequency provided by the HTCsmart mobile phone. All the raw data were stored in a mem-

ory card in the mobile phone as text files. The n-by-3 textarray contained the three axes of acceleration data, left toright: VT acceleration; ML acceleration and AP acceleration,respectively.

web version of the article.)

Firstly, the acceleration data were imported into iGAITusing the ‘Load Signal’ function and all of the three accelera-tion dimensions were displayed on the plot panel as shown inFig. 1. Because the raw data included the data collected beforeand after walking, redundant data were eliminated. This wasachieved by clicking the ‘Region Select’ button (second step),and using the cursor to locate the start and end points whichenclosed the acceleration data. These data were then used inthe following analysis.

Thirdly, the sampling frequency was set as 25 Hz (sampleinterval 40 ms), the walking distance was set as 25 m, and thedefault threshold value (0.4) was used.

When the ‘Gait Detect’ button was clicked, ‘heel contact’points were marked in the AP acceleration plot as black trian-gles. If, through visual inspection, it was apparent that someof these events had been missed they were entered manuallyby using the ‘+’ button. On the other hand the ‘−’ button wasused to remove incorrectly identified events.

The fifth step was to click the ‘Analyse’ button, wherebythe gait features extraction function was called. The autocor-relation coefficient curves of the three direction accelerationdata were displayed in a new window. The local maximumautocorrelation coefficients values around the step time andstride time were automatically marked as a green star anda red star, respectively, on the curves (Fig. 5). In some cases,the local maximum value cannot be automatically found. Inthis instance the mouse can be used to locate the position. Intotal, 31 features were extracted and 18 out of 31 features werepresented on the screen (Fig. 4). The last step was to save thefeatures. The program generated a text file as shown in Fig. 6,

which included all the features calculated by the program.

In order to further demonstrate the application of iGAIT,we applied the iGAIT to analyse the accelerometer data

c o m p u t e r m e t h o d s a n d p r o g r a m s i n b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 715–723 721

F

csctePptIn(btid

4

TsatumtT

Table 1 – The top five features with mean values,standard deviation and p values established by usingindependent unpaired t-test from both control subjectsand CRPS subjects.

Feature Control CRPS p

Mean SD Mean SD

Step regularity VT 0.86 0.07 0.55 0.21 <.01IPSD VT (g2) 0.52 0.29 0.14 0.14 <.01RMS ML (g) 0.16 0.03 0.10 0.03 <.01IPSD AP (g2) 0.39 0.27 0.09 0.02 <.01Stride regularity VT 0.85 0.08 0.62 0.18 <.01

further assessing its practical utility through validation and

ig. 6 – An example of output files generated by iGAIT.

ollected by an accelerometer (Dynaport MiniMod) fromubjects with Complex Region Pain Syndrome (CRPS) andontrol subjects [30]. A total of 21 features including spatio-emporal, frequency, regularity and symmetry features werextracted by iGAIT and were used as input to Multilayererceptron-based classification models to discriminate theatients with CRPS from the control subjects. When usinghe top five features, i.e. Step Regularity VT, IPSD VT, RMS ML,PSD AP and Stride Regularity VT, selected based on signal tooise ratio ranking criterion, the best classification accuracy

97.5%) was achieved. The comparison of the top five featuresetween the control subjects and the subjects with CRPS inerms of their mean values and standard deviation is listedn Table 1. The reader is referred to [30] for a more detailedescription of the application.

. Discussion and summary

his paper presents an interactive, user friendly tool toupport accelerometer based gait data analysis. It can plotcceleration data and allows the user to select portions ofhe data region directly from the plot. Users can also man-ally adjust detection of the gait events, which are under

ost circumstances automatically detected by iGAIT. In total,

he tool can extract 31 gait features from accelerometer data.he results can be displayed on the screen and be saved for

further statistical analysis. A key advantage of the tool isthat it was not developed for specific accelerometer systems.To date it has been tested on three types of accelerometers(Dynaport Minimod, Xsens MT9 and accelerometer embed-ded in HTC mobile) with different sampling frequencies(5–100 Hz).

Initially, iGAIT was designed for lower back accelerationdata. However it is also suitable for data collected on theupper body along the vertebral column, for example the tho-rax, upper back, neck or occipital pole of the head. If anaccelerometer is placed on one side of the upper body, mostgait parameters are still reliably detected excluding the sym-metry features and step regularity features. This is due to theplacement being away from the middle axis of upper body, andresults in the acceleration data referring to the right and leftsteps lacking consistency in all three directions. In many stud-ies accelerometers attached to legs, feet and shoes are usedto measure gait parameters as these sensor locations providethe strongest signals [34–36]. Gait events, such as heel strikeand toe off, can be detected from such accelerometer data.Many parameters related to the lower limb movement can beobtained, for example, swing, single and double limb support.However movement of the trunk cannot be obtained from thelower-limb acceleration data, and multiple sensors have to beused on both sides of the lower limb to obtain gait symmetryfeatures.

Currently the output files of iGait only include the inputdata file name, the feature names and the feature values. Toavoid any confusion, a header will be added to the data andresult files in future versions, which describes the data, suchas, the sensor used, order of channels, sampling rates, sensorlocation, distance walked and user name.

In the future, iGAIT will be evaluated on a larger scale. Forexample, iGAIT is to be tested with acceleration data collectedfrom subjects with different types of syndromes, such as backand neck pain, and pathologies such as Parkinson’s diseaseand stroke. Moreover iGAIT is to be tested with a wider rangeof accelerometer types.

The platform presented in this paper may be inter-preted as a generic framework to facilitate the analysis ofaccelerometer-based gait data. The authors are interested in

integration with other research efforts. Therefore, differentpackages, including an open source version, will be availableupon request from academic users.

s i n

r

722 c o m p u t e r m e t h o d s a n d p r o g r a m

Conflict of interest statement

None of the authors had financial and personal relationshipswith other people or organisations that could inappropriatelyinfluence their work.

Acknowledgements

The work was supported in part by the Vice-Chancellor’sResearch Scholarships (VCRS) from the University of Ulster.We also appreciate the clinical support from Dr. Nigel Harriswho gave us some clinical analysis suggestions.

Appendix A. Supplementary data

Supplementary data associated with this article canbe found, in the online version, at http://dx.doi.org/10.1016/j.cmpb.2012.04.004.

e f e r e n c e s

[1] A. Godfrey, R. Conway, D. Meagher, G. OLaighin, Directmeasurement of human movement by accelerometry,Medical Engineering & Physics 10 (2008) 1364–1386.

[2] W. Zijlstra, K. Aminian, Mobility assessment in older people:new possibilities and challenges, European Journal ofAgeing 1 (2007) 3–12.

[3] W. Zijlstra, Assessment of spatio-temporal parametersduring unconstrained walking, European Journal of AppliedPhysiology 1 (2004) 39–44.

[4] J.J. Kavanagh, S. Morrison, D.A. James, R. Barrett, Reliabilityof segmental accelerations measured using a new wirelessgait analysis system, Journal of Biomechanics 15 (2006)2863–2872.

[5] R. Moe-nilssen, J.L. Helbostad, Estimation of gait cyclecharacteristics by trunk accelerometry, Journal ofBiomedical 1 (2004) 121–126.

[6] B. Auvinet, G. Berrut, C. Touzard, L. Moutel, N. Collet, D.Chaleil, E. Barrey, Reference data for normal subjectsobtained with an accelerometric device, Gait and Posture 2(2002) 124–134.

[7] M. Brandes, W. Zijlstra, S. Heikens, R. van Lummel, D.Rosenbaum, Accelerometry based assessment of gaitparameters in children, Gait and Posture 4 (2006) 482–486.

[8] R. Begg, M. Palaniswami, Computational Intelligence forMovement Sciences: Neural Networks and Other EmergingTechniques, Idea Group Publishing, 2006.

[9] A. Zijlstra, E.D. de Bruin, N. Bruins, W. Zijlstra, The steplength–frequency relationship in physically activecommunity-dwelling older women, European Journal ofApplied Physiology 104 (2008) 427–434.

[10] D. Giansanti, V. Macellari, G. Maccioni, New neural networkclassifier of fall-risk based on the Mahalanobis distance andkinematic parameters assessed by a wearable device,Physiological Measurement 3 (2008) (2008) N11–N19.

[11] L.N. Persch, C. Ugrinowitsch, G. Pereira, A.L.F. Rodacki,Strength training improves fall-related gait kinematics in

the elderly: a randomized controlled trial, ClinicalBiomechanics 10 (2009) 819–825.

[12] J.R. Nocera, T. Buckley, D. Waddell, M.S. Okun, C.J. Hass, Kneeextensor strength, dynamic stability, and functional

b i o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 715–723

ambulation: are they related in Parkinson’s disease? ArchivePhysical Medical Rehabilitation 4 (2010) 589–595.

[13] G. Ebersbach, M. Sojer, F. Valldeoriola, J. Wissel, J. Muller, E.Tolosa, W. Poewe, Comparative analysis of gait inParkinson’s disease, cerebella ataxia and subcorticalarteriosclerotic encephalopathy, Brain 7 (1999) 1349–1355.

[14] Y. Zhou, S. Guo, Q. Liu, J. Tang, Y. Li, Influnce of the femoralhead size on early postoperative gait restoration after totalhip arthroplasty, Chinese Medical Journal 13 (2009)1513–1516.

[15] H. Zheng, N.D. Black, N.D. Harris, Position-sensingtechnologies for movement analysis in stroke rehabilitation,Medical Biology Engineering Computing 4 (2005)413–420.

[16] R. Senden, B. Grimm, I.C. Heyligers, H.H.C.M. Savelberg, K.Meijer, Acceleration-based gait test for healthy subjects:reliability and reference data, Gait and Posture 2 (2009)192–196.

[17] A. Godfrey, K.M. Culhane, G.M. Lyons, Comparison of theperformance of the activPAL (TM) Professional physicalactivity logger to a discrete accelerometer-based activitymonitor, Medical Engineering and Physics 8 (2007)930–934.

[18] A.H. Mackey, N.S. Stott, S.E. Walt, Reliability and validity ofan activity monitor (IDEEA) in the determination oftemporal–spatial gait parameters in individuals withcerebral palsy, Gait and Posture 4 (2008) 634–639.

[19] N.A. Maffiuletti, M. Gorelick, I.K. Quervain, M. Bizzini, J.P.Munzinger, S. Tomasetti, A. Stacoff, Concurrent validity andintrasession reliability of the IDEEA accelerometry systemfor the quantification of spatiotemporal gait parameters,Gait and Posture 1 (2008) 160–163.

[20] G. Pendharkar, S. Marin, R. Mayagoita-Hill, P. Percival,Comparing the head and trunk accelerations during a freewalk using MTx sensors, Intelligent Sensors, SensorNetworks and Information (2007) 631–634.

[21] K. Turcot, R. Aissaoui, K. Boivin, M. Pelletier, N. Hagemeister,J.A. de Guise, New accelerometric method to discriminatebetween asymptomatic subjects and patients with medialknee osteoarthritis during 3D gait, IEEE Transactions onBiomedical Engineering 4 (2008) 1415–1422.

[22] A. Tura, M. Raggi, L. Rocchi, A.G. Cutti, L. Chiari, Gaitsymmetry and regularity in transfemoral amputeesassessed by trunk accelerations, Journal ofNeuroengineering and Rehabilitation 7 (2010) 4.

[23] Y.L. Kuo, K.M. Culhane, P. Thomason, O. Tirosh, R. Baker,Measuring distance walked and step count in children withcerebral palsy: an evaluation of two portable activitymonitors, Gait and Posture 2 (2009) 304–310.

[24] R. Moe-Nilssen, A new method for evaluating motor controlin gait under real-life environmental conditions. Part 2: Gaitanalysis, Clinical Biomechanics 13 (1998) 328–335.

[25] DynaPort GaitTest, Available from: http://www.mcroberts.nl/products/clinician/dynaport-gaittest (accessed 24.08.10).

[26] M. Hausdorff, Gait dynamics, fractals and falls: findingmeaning in the stride-to-stride fluctuations of humanwalking, Human Movment Science 4 (2007)555–589.

[27] D. Roberts, R. Hamill-Ruth, B. Parker, S. Maximous, B. Clark,K. Nelson, A new method for evaluating efficacy of low backpain therapy, Journal of Pain 3 (2004) S116.

[28] C.J. Lamoth, J.F. Stins, M. Pont, F. Kerckhoff, P.J. Beek, Effectsof attention on the control of locomotion in individuals withchronic low back pain, Journal of Neuroengineering andRehabilitation 5 (2008) 13.

[29] D. Newell, M. van der Laan, The fractal dynamics of walkingin chronic non specific low back pain patients, ClinicalChiropractic 1 (2010) 8–14.

i n b i

c o m p u t e r m e t h o d s a n d p r o g r a m s

[30] M. Yang, H. Zheng, H. Wang, S. McClean, J. Hall, N. Haris,Assessing accelerometer based gait features to support gaitanalysis for people with complex regional pain syndrome,in: Proceedings of the 3rd International Conference onPervasive Technologies Related to Assistive Environments,Samos, Greece, 2010.

[31] A. Goldberger, L. Amaral, J. Hausdorff, P. Ivanov, C. Peng, H.Stanley, Fractal dynamics in physiology: alterations withdisease and ageing, Proceeding of National AcademicScience 1 (2002) 2466–2472.

[32] M.N. Nyan, F.E.H. Tay, K.H.W. Seah, Y.Y. Sitoh, Classification

of gait patterns in the time–frequency domain, Journal ofBiomechanics (2006) 2647–2656.

[33] P. Schwingenschuh, E. Katschnig, C. Ott, Enzinger, Gaitanalysis in patients with Parkinson’s disease off

o m e d i c i n e 1 0 8 ( 2 0 1 2 ) 715–723 723

dopaminergic therapy, Archive Physical MedicalRehabilitation 11 (2009) 1880–1886.

[34] J.D. Schaafsma, N. Giladi, Y. Balash, A.L. Bartels, T. Gurevich,J.M. Hausdorff, Gait dynamics in Parkinson’s disease:relationship to Parkinsonian features, falls and response tolevodopa, Journal of Neurological Sciences 1–2 (2003)47–53.

[35] S.T. Moore, H.G. MacDougall, W.G. Ondo, Ambulatorymonitoring of freezing of gait in Parkinson’s disease, Journalof Neuroscience Methods 167 (2008) 340–348.

[36] K. Saremi, J. Marehbian, X.H. Yan, J.P. Regnaux, R. Elashoff, B.

Bussel, B.H. Dobkin, Reliability and validity of bilateral thighand foot accelerometry measures of walking in healthy andhemiparetic subjects, Neurorehabilitation and Neural Repair20 (2006) 297–305.