A WEARABLE GAIT ANALYSIS SYSTEM USING INERTIAL

SENSORS PART II

Evaluation in a Clinical Setting

A. Sant’Anna

, N. Wickstr

H. Eklund

and R. Tranberg

Intelligent Systems Lab, Halmstad University, Halmstad, Sweden

Center for Person-Centered Care, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden

Department of Orthopedics, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden

Keywords:

Gait Analysis, Inertial Sensors, Symmetry, Normality, Clinical Environment.

Abstract:

The gold standard for gait analysis, in-lab 3D motion capture, is not routinely used for clinical assessment due

to limitations in availability, cost and required training. Inexpensive alternatives to quantitative gait analysis

are needed to increase the its adoption. Inertial sensors such as accelerometers and gyroscopes are promising

tools for the development of wearable gait analysis (WGA) systems. The present study evaluates the use of

a WGA system on hip-arthroplasty patients in a real clinical setting. The system provides information about

gait symmetry and normality. Results show that the normality measurements are well correlated with various

quantitative and qualitative measures of recovery and health status.

1 INTRODUCTION

Gait analysis (GA) is a tool that can aid the assess-

ment of several physical and cognitive conditions.

Perhaps the most widely adopted use of GA is in

the treatment of cerebral palsy children (Chang et al.,

2010), (DeLuca et al., 1997). A few other areas

have also investigated GA as an aid to clinical assess-

ment, e.g. Parkinson’s disease (Salarian et al., 2004),

(Frenkel-Toledo et al., 2005), and stroke (Cruz and

Dhaher, 2008), (Silver et al., 2000). Despite many

positive results, GA is still not routinely used in the

clinical setting.

Several factors contribute to the low adoption of

GA as a clinical tool. The gold standard for GA is 3D

in-lab motion capture (MOCAP), which can consider-

ably improve clinical assessment, e.g. (Lofterød and

Terjesen, 2008), (Morais Filho et al., 2008). However,

the interpretation of results requires speciﬁc training

and experience, and its use as a routine procedure is

not widely accepted from an economical perspective

(Simon, 2004). In addition, a state-of-the-art MO-

CAP system is often not available in small practices

and underpreviledged areas. The alternative to MO-

CAP is observational gait analysis (OGA), which is

intrinsically subjective and sensitive to the observer’s

experience (Toro et al., 2003). Nonetheless, in 1999

Coutts (Coutts, 1999) argued that despite its limita-

tions, OGA would never be totally replaced as the de-

fault GA method in the clinical environment because

of ease of use.

Recently large efforts have been employed in de-

veloping low-cost, mobile GA systems that can com-

plement OGA with objective, reliable, quantitative

data. The success of such systems will hopefully in-

cur in a wide-spread adoption of quantitative GA as

a clinical tool. The design of these system may be

guided by a few practical requirements:

• Easy to use: It is important that the system require

no special training and very little extra effort from

the staff in order to ensure cost effectiveness. It is

also desirable that the system be used quickly and

independent of infrastructure/location.

• Unobtrusive: To spare patients, who may be in

pain or discomfort, the system must be small, light

weight, and comfortable to wear. In addition, the

system should not interfere with the patient’s nat-

ural movements. Cumbersome equipment may

make the patient feel self conscious, and act un-

naturally.

• Valid and reliable: The information acquired by

the system must be accurate, and representative

of clinical assessment.

• Easy to interpret: Kinematic information and

other speciﬁc gait parameters are mostly useful

Sant’Anna A., Wickström N., Eklund H. and Tranberg R..

A WEARABLE GAIT ANALYSIS SYSTEM USING INERTIAL SENSORS PART II - Evaluation in a Clinical Setting.

DOI: 10.5220/0003707700050014

In Proceedings of the International Conference on Bio-inspired Systems and Signal Processing (BIOSIGNALS-2012), pages 5-14

ISBN: 978-989-8425-89-8

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

for clinicians who have experience with gait anal-

ysis. To ensure wide-spread adoption, the system

must provide intuitive and easy to interpret infor-

mation.

The present study aims at developing a small

wearable system that fulﬁlls the above requirements,

and can be successfully deployed in the clinical set-

ting. Part I of the study introduced a symbolic ap-

proach to the analysis of gait symmetry and normal-

ity using miniature accelerometers and gyroscopes.

The proposed method was then compared to symme-

try and normality measures extracted from 3D MO-

CAP kinematic data.

Part II now investigates the viability of using the

system to evaluate patients in a real clinical setting.

The data collection took place at the orthopedic ward

at Sahlgrenska Univeristy Hospital, M

olndal, Swe-

den. Eleven unilateral hip-arthroplasty patients un-

derwent GA with the proposed system at discharge,

and once again approximately three months later.

Measures of symmetry and normality were derived

and evaluated against a timed 10-meter walk test and

a EQ-5D health questionnaire.

2 RELATED WORK

Due to limitations in availability, cost and training

required for 3D GA, this section only discusses GA

methods that can be performed inexpensively and in-

dependently of a MOCAP gait lab, namely observa-

tional gait analysis (OGA) and wearable sensor sys-

tems.

2.1 Observational Gait Analysis

Observational gait analysis (OGA) can be further di-

vided into: naked eye observation (NE-OGA), and

video-aided observation (VA-OGA). VA-OGA has a

clear advantage over NE-OGA in that it allows more

freedom to the observer, enabling pause, slow motion,

and other functions. In some cases, quantitative mea-

surements, such as joint angles (Embrey et al., 1990),

can be directly calculated from the image.

VA-OGA is often accompanied by a form or ques-

tionnaire that facilitates the extraction of relevant in-

formation from the video. These forms typically em-

ploy binary or gross estimates of variables, such as

Presence/Absence or Normal/Mild/Sever. Very fre-

quently, new questionnaires are developed for speciﬁc

studies or clinics, e.g. (Brunnekreef et al., 2005),

(Kawamura et al., 2007). Although various ques-

tionnaires evaluate similar features, they often differ

largely in rating scheme and variables of interest.

Two VA-OGA forms have been more thoroughly

investigated and more widely adopted: the Visual

Gait Assessment Scale (VGAS) (Dickens and Smith,

2006), (Brown et al., 2008) and the Edinburgh Visual

Gait Score (EVGS) (Read et al., 2003), (Ong et al.,

2008). Both questionnaires target the assessment of

ambulatory children with cerebral palsy. Form and

questionnaires have also been used to aid NE-OGA.

The physician Rating Scale seems to be the most fre-

quently used, (Koman et al., 1993), (Maathuis et al.,

2005), (Wren et al., 2005), also targeted to children

with cerebral palsy.

OGA can be complemented by other more quan-

titative measurements, such as average gait speed, av-

erage step length and other gait parameters. These are

typically measured during walking tests, such as the

10-m walking test (Dean et al., 2001), (Kempen et al.,

2011), or the timed up and go test (TUG) (Kristensen

et al., 2007), (Nordin et al., 2008). The TUG is nor-

mally employed in studies where balance and risk of

fall are of interest, as it requires that the subject stand

up and sit down on a chair without help.

The 10-m walking test, on the other hand, is a

simple way of determining, average gait speed, stride

length and cadence. Average gait speed, for exam-

ple, has been identiﬁed as an indicator of: activity

of daily living function in geriatric patients (Potter

et al., 1995); high risk of health-related outcomes in

well-functioning older people (Cesari et al., 2005),

and leg strength in older people (Buchner et al.,

1996). Stride length is another interesting measure

that has been associated with, for example, metabolic

cost and impact during walking (Russell et al., 2010),

(Mercer et al., 2003).

2.2 Mobile GA Systems

Current sensor and hardware technologies have made

possible the creation of small wearable systems for

GA. A large number of authors have dedicated their

efforts to developing mobile, simpler alternatives to

3D in-lab GA. Such systems may be categorized

according to the information they produce: spatio-

temporal (ST) parameters, kinematics, or overall

characteristics of gait. These categories are not neces-

sarily mutually exclusive, but they represent different

levels of information complexity.

ST parameters, such as stride time and veloc-

ity, can only convey information about when and/or

where the foot hits the ground. One way to visualize

this is to imagine that ST parameters could be cal-

culated from a series of foot-prints over time. Sys-

tems that measure ST parameters are usually simpler

and more commonly used. Some of the earliest sys-

BIOSIGNALS 2012 - International Conference on Bio-inspired Systems and Signal Processing

tems employed foot-switches to determine initial and

terminal contact, e.g. (Hausdorff and Ladin, 1995).

More recently, studies have found accelerometer mea-

surements valid and reliable means of determining

walking speed, cadence, stride length and other ST

parameters, e.g. (Saremi et al., 2006), (Senden et al.,

2009), (Bautmans et al., 2011). Although ST infor-

mation can be very useful, it does not represent the

subject’s gait pattern as a whole. It is important to

know what happens between foot-prints.

The second category encompasses those systems

that are able to extract kinematic data such as trajec-

tories, and joint angles. Some of these systems pro-

vide only foot pitch and ground incline in addition to

ST information, e.g. (Sabatini et al., 2005), (Bam-

berg et al., 2008). Others provide measures of joint

angle progressions, segment rotations and accelera-

tions, e.g. (Dejnabadi et al., 2005), (Mayagoitia et al.,

2002b). These systems can provide an inexpensive

alternative to in-lab 3D GA. However, proper training

and experience are required for interpreting kinematic

information. In addition, these systems are too cum-

bersome to be used for extended periods of times.

The third category aims at extracting more gen-

eral characteristics of gait such as gait symmetry

(Gouwanda and Senanayake, 2011), gait regularity

(Moe-Nilssen and Helbostad, 2004), and balance (Al-

lum and Carpenter, 2005), (Mayagoitia et al., 2002a).

Although this information may be derived from ST

and/or kinematic data, systems can be made much

simpler if they directly measure general character-

istics. For example, (Moe-Nilssen and Helbostad,

2004) measures gait symmetry using only one ac-

celerometer placed on the lower back, whereas all

methods mentioned in the previous categories make

use of at least one sensor node on each lower limb.

General characteristics of gait are usually not enough

for determining the cause of the subject’s gait abnor-

mality. However, they are easy to interpret and can be

used to monitor the subject’s progress after treatment

has been established.

Although measures of symmetry have been

largely investigated, measures of gait normality are

lacking. In Part I of the present study a measure

of normality was introduced, derived from three sen-

sor nodes containing accelerometers and gyroscopes.

This measure was compared to measures of normal-

ity computed from 3D kinematic data. In Part II of

the study, the viability and usefulness of the proposed

normality index is investigated in a clinical setting.

3 METHOD

3.1 Data Collection

Eleven patients were included in the study. All pa-

tients had undergone unilateral hip-arthroplasty for

the ﬁrst time and presented no other physical or cog-

nitive conditions. The group was composed of four

women and seven men, the mean age was 69±15

years, mean weight was 81±20 Kg, and mean height

was 172±9 cm.

The data collection was designed to be very quick

in consideration of the patients, most of whom were

in great discomfort. The patients were equipped with

three Shimmer



sensor nodes. One node was placed

on each outer shank, approximately three centimeters

above the lateral malleolus, the remaining node was

placed directly under the navel. Sensors were secured

on the skin with surgical tape. The sensor nodes were

synchronized with the help of beacon signals from a

host computer and the data was stored on-board in a

micro SD card.

The patients were then asked to walk by them-

selves along a 10-meter walkway at a comfortable

speed, twice. The walkway was marked with black

tape on the ﬂoor. The time and number of steps taken

to complete the walkway were recorded at each time.

This procedure took place on the day the patient

was discharged from the hospital, and a few months

later, when the patient came back for a follow-up eval-

uation. The average number of days spent at the ward

after surgery was 4±1 day. The time between base-

line and follow-up measurements was 108±15 days.

All patients employed a walking aid during baseline

measurements, six used two crutches and ﬁve used a

walker with wheels. During follow-up measurements

six patients used one crutch and ﬁve patients walked

without any aiding device.

Patients ﬁlled out an EQ-5D

health question-

naire (Swedish version) approximately two weeks be-

fore surgery, and soon after their follow-up session.

The EQ-5D

is a standardized instrument for use as

a measure of health outcome, developed by the Euro-

Qol Group (www.euroqol.org). The English version

of the questionnaire, validated for Ireland, is shown

in Figures 1 and 2. Each answer is given a value from

1 to 3, good health state results in low values.

This study was approved by the Regional Ethics

Board in Gothenburg, Sweden.

3.2 Observational Gait Analysis

The time, Tm, and number of steps, NumSteps, taken

to complete the 10-meter walk test were used to com-

A WEARABLE GAIT ANALYSIS SYSTEM USING INERTIAL SENSORS PART II - Evaluation in a Clinical Setting

By placing a tick in one box in each group below, please

indicate which statements best describe your own health

state today.

Mobility

I have no problems in walking about 1

I have some problems in walking about 2

I am conﬁned to bed 3

Self-Care

I have no problems with self-care 1

I have some problems washing or dressing 2

I am unable to wash or dress myself 3

Usual Activities (e.g. work, study, house-

work, family or leisure activities)

I have no problems with performing my usual

activities

I have some problems with performing my

usual activities

I am unable to perform my usual activities 3

Pain/Discomfort

I have no pain or discomfort 1

I have moderate pain or discomfort 2

I have extreme pain or discomfort 3

Anxiety/Depression

I am not anxious or depressed 1

I am moderately anxious or depressed 2

I am extremely anxious or depressed 3

Figure 1: EQ-5D

Part A: English version validated for

Ireland.

1990 EuroQol Group EQ-5D

is a trademark

of the EuroQol Group.

pute average speed, Speed = 10/T m (m/s), and av-

erage step length, StepLeng = 10/NumSteps (m). In

addition, step length was normalized by the patient’s

height. These variables were used as reference for the

improvement of the patient, under the assumption that

average speed and step length should increase as the

patient recovers.

3.3 Inertial Sensor Gait Analysis

The method used to compute symmetry and normal-

ity from the sensor data is described in more detail in

Part I of this study. The sensor data is ﬁrst symbol-

ized into equiprobable symbols. The periods between

consecutive similar symbols are calculated and repre-

sented as period histograms. Symmetry is calculated

from the gyroscope data of the shanks as the relative

error between the histograms derived from the right

and left sides. Similarly, Normality is derived from

the accelerometer data of the waist as the difference

between the subject’s histograms and histograms de-

rived from a reference set of healthy individuals. The

reference data set was obtained from healthy, con-

To help people say how good or bad a health

state is, we have drown a scale (rather like

a thermometer) on which the best state you

can imagine is marked 100 and the worst

state you can imagine is marked 0.

We would like you to indicate on this scale

how good or bad your own health is today,

in your opinion. Please do this by drawing a

line from the box below to whichever point

on the scale indicates how good or bad your

health state is today.

Figure 2: EQ-5D

Part B: English version validated for

Ireland.

1990 EuroQol Group EQ-5D

is a trademark

of the EuroQol Group.

siderably younger individuals, average age 34 ± 13

years, walking normally.

The histograms are normalized to stride time so

as to be independent of stride frequency. This nor-

malization by stride variables is common in kinematic

gait analysis and it ensures that results are not biased

by gait cadence, stride length or speed.

During the follow-up session, the gait of the pa-

tients may be considerably different from those of

the reference group. Nonetheless, it is expected that

patients’ gait patterns approach the reference group

more at follow-up than at baseline.

3.4 Data Analysis

The Spearman’s rank correlation coefﬁcient was used

to evaluate the correlation between two variables. The

non-parametric Wilcoxon rank sum test was used to

compare two distributions, and a Kruskal-Wallis test

was used to compare more than two distributions. All

linear model approximations were calculated based

on least mean square errors.

The area under the receiver operating character-

istic curve (AUC) was used to evaluate the discrimi-

natory power of the normality index. The ROC curve

was constructed based on tests performed on the same

individuals. Therefore, any statistically signiﬁcant

comparison between different AUC must take into ac-

count the correlated nature of the data. A nonpara-

BIOSIGNALS 2012 - International Conference on Bio-inspired Systems and Signal Processing

metric approach based on generalized U-statistics was

used to estimate the covariance matrix of the different

curves (DeLong et al., 1988).

All measurements included two trials, which were

used to assess the test-retest reliability of each index

using intra-class correlation coefﬁcient (ICC) type A-

1 as a measure of absolute agreement (McGraw and

Wong, 1996). All tests were bi-directional with con-

ﬁdence level, α = 0.05. All data analysis was under-

taken in MATLAB (MathWorks, Natick, MA).

4 RESULTS

All but one participant answered the EQ-5D

ques-

tionnaire on both occasions. The values of the an-

swers given to each category were added to a single

score for that category. Results from before the op-

eration and after the follow-up session are shown in

Figure 3. Lower scores correspond to more patients

in better health. The biggest changes were regarding

mobility, usual activities and pain/discomfort.

mobility self-care activities pain/disc. anxiety/depr.

sum of questionnaire scores

before surgery

after follow-up

Figure 3: Questionnaire results from before the surgery and

after the follow-up sessions. Lower scores correspond to

more patients in better health.

Symmetry results for baseline and follow-up mea-

surements are shown in Figure 4 for each subject.

Measurements were averaged over both trials of each

session. The symmetry index ranges from 0 to

100, a low symmetry index indicates good symmetry

whereas a high value indicates asymmetry. Accord-

ing to the proposed index, gait symmetry improved at

follow-up for approximately half the subjects. The

asymmetry at follow-up may be caused by the use

of one crutch. The symmetry index according walk-

ing aid is shown in Figure 5. There is a clear dif-

ference between the symmetry of patients using two

crutches at baseline and patients walking with no aid

at follow-up. However, none of the distributions were

signiﬁcantly different.

baseline follow-up

symmetry - shank sensors

subj1

subj2

subj3

subj4

subj5

subj6

subj7

subj8

subj9

subj10

subj11

Figure 4: Symmetry results at baseline and follow-up. The

two distributions are not statistically signiﬁcantly different.

walker crutches one crutch none

symmetry - shank sensors

baseline - follow-up

Figure 5: Symmetry results according to walking aid. Be-

sides the data points, the plot presents box-plot representa-

tions of the distributions. The whiskers represent the small-

est and largest observations, the edges of the box correspond

to the lower and upper quartiles, the horizontal line indi-

cates the median. The distributions are not statistically sig-

niﬁcantly different.

Normality results are shown in Figure 6, measure-

ments for each patient were averaged over both tri-

als of each session. Similarly, the normality index

ranges between 0 and 100, and a low value indicates

good normality. In this case, the follow-up measure-

ments were better than baseline measurements for all

patients. A Wilcoxon test indicated that baseline and

follow-up groups were statistically signiﬁcantly dif-

ferent, p<0.0001. Figure 7 illustrates the distribution

of the normality index according to walking aid. As

expected, the normality index for those patients walk-

ing without aid was, on average, better than the others.

A Kruskal-Wallis test indicated that the free walking

group was statistically different from the walker and

crutches groups, and that the one crutch group was

statistically different from the walker group.

In order to calculate the correlation between nor-

mality and walking aid, each category was repre-

sented by a number. In the order shown in Figure 7,

walker was represented by 1 and no-aid was repre-

A WEARABLE GAIT ANALYSIS SYSTEM USING INERTIAL SENSORS PART II - Evaluation in a Clinical Setting

baseline follow-up

100

normality - waist sensor

subj1

subj2

subj3

subj4

subj5

subj6

subj7

subj8

subj9

subj10

subj11

Figure 6: Normality results at baseline and at follow-up.

The two distributions are statistically different according to

a Wilcoxon rank sum test, p<0.0001.

walker crutches one crutch none

100

normality - waist sensor

baseline - follow-up

Figure 7: Normality results according to walking aid. A

Kruskal-Wallis test indicates that the distribution of no

walking aid is signiﬁcantly different from distributions of

two crutches and walker. The one crutch distribution is sig-

niﬁcantly different from the walker distribution. Normality

is well correlated with walking aid, according to a Spear-

man’s rank correlation coefﬁcient of r=-0.78, p<0.0001.

sented by 4. The resulting Spearman’s rank correla-

tion coefﬁcient was r=-0.78, p<0.0001.

The normality index also correlates well with both

average speed, r=-0.79 p<0.0001, and normalized av-

erage step length, r=-0.76 p<0.0001. Normality val-

ues for each individual trial are shown against aver-

age speed values in Figure 8, and against normal-

ized step length in Figure 9. On both plots the lin-

ear model approximation is shown as a solid line, and

the 95% conﬁdence interval (CI) for predicted obser-

vations is shown as dotted lines. The mean average

speed at baseline, 0.46±0.16 m/s, was signiﬁcantly

different from the speed at follow-up, 1.06±0.22 m/s,

p<0.0001.

Normality results were also compared to the EQ-

answers that varied the most between be-

fore the surgery and after follow-up, namely mo-

bility (Figure 10), usual activities (Figure 11), and

pain/discomfort (Figure 10). In all cases, there is a

0 0.5 1

100

105

average speed (m/s) - reference data

normality index - waist sensor

fitted curve

lower bound 95% CI

upper bound 95% CI

baseline trial1

baseline trial2

follow-up trial1

follow-up trial2

Figure 8: Normality compared to average speed. Variables

are well correlated, Spearman’s rank correlation coefﬁcient

r=-0.79, p<0.0001. The solid line indicates the linear model

approximation a+bx, where a=95.5 with conﬁdence inter-

val (CI) [94.9, 96.3]; and b=-2.6 with CI [-3.3, -1.9]. The

dashed and dotted lines indicate the 95% CI of predicted

observations.

10 20 30 40 50

100

105

normalized step length (% of hight)

normality index - waist sensor

fitted curve

lower bound 95% CI

upper bound 95% CI

baseline trial1

baseline trial2

follow-up trial1

follow-up trial2

Figure 9: Normality compared to average step length. Vari-

ables are well correlated, Spearman’s rank correlation co-

efﬁcient r=-0.76, p<0.0001. The solid line indicates the

linear model approximation a+bx, where a=95.5 with con-

ﬁdence interval (CI) [94.9, 96.3]; and b=-2.5 with CI [-3.2,

-1.8]. The dashed and dotted lines indicate the 95% CI of

predicted observations.

tendency for better health to be accompanied by bet-

ter normality index. This correlation is particularly

strong between normality and usual activities scores,

Spearman’s r=0.75, p=0.0127. There was no correla-

tion between the health scale in Part B of the ques-

tionnaire and normality.

Improvement in normality was calculated as the

difference between baseline and follow-up values.

Figure 12 shows how improvement in normality cor-

relates with number of days spent at the ward af-

ter surgery. Although a Wilcoxon test indicated that

there was no statistically signiﬁcant difference be-

BIOSIGNALS 2012 - International Conference on Bio-inspired Systems and Signal Processing

ans 1 ans 2

100

normality - waist sensor

A) EQ-5D mobility

ans 1 ans 2

100

normality - waist sensor

B) EQ-5D pain/discomfort

Figure 10: Normality compared to EQ-5D

answers re-

garding (A) mobility and (B) pain/discomfort. Mobility an-

swers - ans 1: I have no problems in walking about; ans 2: I

have some problems in walking about. Pain/discomfort an-

swers - ans 1: I have no pain or discomfort; ans 2: I have

moderate pain or discomfort.

ans 1 ans 2 ans 3

100

normality - waist sensor

EQ-5D usual activities

Figure 11: Normality compared to EQ-5D

answers re-

garding usual activities. Ans 1: I have no problems with

performing my usual activities; ans 2: I have some problems

with performing my usual activities; ans 3: I am unable to

perform my usual activities. Variables are well correlated,

Spearman’s rank correlation coefﬁcient r=0.75, p=0.0127.

tween groups, the Spearman’s rank correlation coef-

ﬁcient was r=-0.75, p=0.0081. There was no correla-

tion between improvement in normality and days be-

tween baseline and follow-up sessions.

The normality index can also be evaluated based

on its discriminatory values. That is, the ability to

differentiate baseline measurements from follow-up

measurements. The AUC was 0.94, conﬁdence in-

terval (CI) (0.87, 1.00), p<0.0001. The test-retest

reliability was also high, r=0.81, CI (0.60, 0.92),

p<0.0001.

5 DISCUSSION

The average speeds at baseline and follow-up are in

agreement with measurements reported in (Kennedy

2 days 3 days 4 days 6 days

Improvement in normality

days at the ward after surgery

subj5

subj10

subj2

subj9

subj11

subj3

subj4

subj6

subj7

subj8

subj1

Figure 12: Improvement in normality compared to days

spent at ward after surgery. Improvement in normality is

the difference between normality values at baseline and at

follow-up. Although the distributions are not statistically

different, variables are well correlated, Spearman’s rank

correlation coefﬁcient r=-0.75, p=0.0081.

et al., 2005), 0.46 m/s less than 16 days after hip re-

placement surgery and 1.17 m/s more than 20 days

after surgery. Average gait speed of approximately

1 m/s three months after surgery were also reported

in (Aminian et al., 1999). According to (Macni-

col et al., 1980) the greatest improvements in gait

speed are observed within the ﬁrst three months post-

op. The follow-up measurement can, therefore, be

considered representative of patient’s improvement in

gait speed. In addition, (Palombaro et al., 2006) deter-

mined that changes in speed superior to 0.10 m/s are

clinically meaningful after hip fracture treatment. The

changes in speed observed from baseline to follow-

up, 0.60±0.29, are therefore also clinically meaning-

ful.

Measures of gait normality correlate well with

both gait speed, Figure 8 and step length, Figure 9.

Given that speed and step length are measures related

to patient recovery, there is a good chance the normal-

ity index is also a good indicator of recovery. Unfor-

tunately, no other quantitative gait parameters were

available in the data set to demonstrate that the nor-

mality index correlates to recovery when the data is

corrected for speed. However, in Part I of this study

symmetry and normality measures are shown to cor-

relate to joint-angle curves normalized to stride time,

not containing any velocity information. The normal-

ity index is also normalized to stride time and as such

is independent of walking cadence. It is expected that

the normality index would differentiate between nor-

mal and abnormal patterns at the same speed. Further

investigations are needed to support this assumption.

Another factor supporting the usefulness of the

normality index is its correlation with the type of

walking aid used during the test, Figure 7. The test-

retest reliability and discriminatory power of the in-

A WEARABLE GAIT ANALYSIS SYSTEM USING INERTIAL SENSORS PART II - Evaluation in a Clinical Setting

dex were also satisfactory. Overall, the proposed in-

dex can possibly be developed into a reliable and clin-

ically relevant measure of gait normality.

Another interesting result was the correlation be-

tween improvement of normality and number of days

spent at the ward, Figure 12. Whereas there was no

correlation between improvement in normality and

number of days between baseline and follow-up. This

possibly suggests that the rate of recovery at the ward

is indicative of the total rate of recovery, which is little

affected by the recovery time at home. This assump-

tion should be further investigated.

Normality results and the answers to the EQ-5D

questionnaire showed some positive trends. Greater

discomfort and difﬁculties in performing usual activ-

ities seem to be accompanied with worse normality,

Figure 10. Besides the self-assessment questionnaire,

the use of walking aids was also considered an indi-

cation of how well the patient’s health status was, i.e.

patients who did not need any walking aid were, on

average, in better condition than those who used one

crutch. Another indicator of recovery was the num-

ber of days the patient spent at ward, assuming that

patients who recovered better or more quickly were

discharged sooner. The normality index seems to be

in agreement with all the above mentioned qualitative

health status assessments.

Symmetry results are difﬁcult to judge due to the

variety of walking aids used. The large variety of

symmetry at follow-up, Figure 4, was mostly inﬂu-

enced by the patients using one crutch only. This

could be explained by the fact that some patients

were more dependent on the crutch and consequently

leaned more to one side. Whereas some patients

barely used the crutch for support.

Due to their recent surgery, patients were very un-

comfortable during the baseline measurements. It was

important to keep the data collection as simple and

quick as possible. No more than ﬁve minutes had to

be spared by the patient to complete the entire pro-

cedure, and they were all willing to participate in the

study. Briefness is also important for the staff respon-

sible for the procedure so that the addition of GA is

not an extra burden. The placement of the sensors was

also quick and easy. However, in the future, the waist

sensor should be placed on the lower back so as not

to be affected by subjects’ different shapes and sizes.

Another issue with the present study is that the

number of participants was very small. Any statistical

inference on the results is greatly affected by the sam-

ple size. However, results are promising and suggest

that a larger study will likely produce positive results.

At the ward where the data was collected, gait

analysis is not normally used, and most records are

based on rough qualitative descriptions. This lack

of quantitative measures makes the assessment of pa-

tient improvement a difﬁcult and very subjective task.

The introduction of a simple 10-meter walk test can

already provide quantitative measures of speed and

stride length. The addition of a wearable GA system,

however, can quickly increase the amount of quanti-

tative data to include more complex measures of sym-

metry and normality.

6 CONCLUSIONS

The present study investigated the viability of using a

wearable gait analysis (GA) system to assess patients

in a real clinical environment. The proposed system

served as a tool to facilitate the extraction of certain

gait characteristics, namely symmetry and normality.

The system was easy to use and did not require more

than ﬁve minutes to complete the entire test proce-

dure. It was small, light weight, and did not interfere

with the patient’s movements. In addition, the infor-

mation provided by the system correlated well with

the level of recovery and health status of the patients

in a very intuitive way. The proposed system, there-

fore, fulﬁlls the practical requirements that are essen-

tial to the successful implementation of wearable GA

systems as clinical tools.

ACKNOWLEDGMENTS

This study was partially funded by the Promobilia

Foundation. The authors would like to thank Lars-

Eric Olsson, PhD, for including the present study

in his project, funded by the Institute of Health and

Care Sciences, Sahlgrenska Academy, University of

Gothenburg, Sweden.

REFERENCES

Allum, J. H. J. and Carpenter, M. G. (2005). A speedy so-

lution for balance and gait analysis: angular velocity

measured at the centre of body mass. Current Opinion

in Neurology, 18(1):15–21.

Aminian, K., Rezakhanlou, K., De Andres, E., Fritsch, C.,

Leyvraz, P. F., and Robert, P. (1999). Temporal feature

estimation during walking using miniature accelerom-

eters: an analysis of gait improvement after hip arthro-

plasty. Medical and Biological Engineering and Com-

puting, 37:686–691.

Bamberg, S., Benbasat, A., Scarborough, D., Krebs, D.,

and Paradiso, J. (2008). Gait analysis using a shoe-

integrated wireless sensor system. IEEE Transactions

BIOSIGNALS 2012 - International Conference on Bio-inspired Systems and Signal Processing

on Information Technology in Biomedicine, 12(4):413

–423.

Bautmans, I., Jansen, B., Keymolen, B. V., and Mets,

T. (2011). Reliability and clinical correlates of 3d-

accelerometry based gait analysis outcomes according

to age and fall-risk. Gait & Posture, 33(3):366 – 372.

Brown, C., Hillman, S., Richardson, A., Herman, J., and

Robb, J. (2008). Reliability and validity of the visual

gait assessment scale for children with hemiplegic

cerebral palsy when used by experienced and inexpe-

rienced observers. Gait & Posture, 27(4):648–652.

Brunnekreef, J., van Uden, C., van Moorsel, S., and

Kooloos, J. (2005). Reliability of videotaped obser-

vational gait analysis in patients with orthopedic im-

pairments. BMC Musculoskeletal Disorders, 6(1):17.

Buchner, D. M., Larson, E. B., Wagner, E. H., Koepsell,

T. D., and De Lateur, B. J. (1996). Evidence for a

non-linear relationship between leg strength and gait

speed. Age and Ageing, 25(5):386–391.

Cesari, M., Kritchevsky, S. B., Penninx, B. W. H. J., Nick-

las, B. J., Simonsick, E. M., Newman, A. B., Tylavsky,

F. A., Brach, J. S., Satterﬁeld, S., Bauer, D. C.,

Visser, M., Rubin, S. M., Harris, T. B., and Pahor, M.

(2005). Prognostic value of usual gait speed in well-

functioning older people - results from the health, ag-

ing and body composition study. Journal of the Amer-

ican Geriatrics Society, 53(10):1675–1680.

Chang, F. M., Rhodes, J. T., Flynn, K. M., and Carollo,

J. J. (2010). The role of gait analysis in treating gait

abnormalities in cerebral palsy. Orthopedic Clinics of

North America, 41(4):489 – 506.

Coutts, F. (1999). Gait analysis in the therapeutic environ-

ment. Manual Therapy, 4(1):2 – 10.

Cruz, T. H. and Dhaher, Y. Y. (2008). Evidence of abnormal

lower-limb torque coupling after stroke: An isomet-

ric study supplemental materials and methods. Stroke,

39(1):139–147.

Dean, C. M., Richards, C. L., and Malouin, F. (2001).

Walking speed over 10 metres overestimates loco-

motor capacity after stroke. Clinical Rehabilitation,

15(4):415–421.

Dejnabadi, H., Jolles, B., and Aminian, K. (2005). A new

approach to accurate measurement of uniaxial joint

angles based on a combination of accelerometers and

gyroscopes. IEEE Transactions on Biomedical Engi-

neering, 52(8):1478 –1484.

DeLong, E. R., DeLong, D. M., and Clarke-Pearson, D. L.

(1988). Comparing the areas under two or more corre-

lated receiver operating characteristic curves: A non-

parametric approach. Biometrics, 44(3):837–845.

DeLuca, P. A., Davis, R. B.,

Ounpuu, S., Rose, S., and

Sirkin, R. (1997). Alterations in surgical decision

making in patients with cerebral palsy based on three-

dimensional gait analysis. Journal of Pediatric Or-

thopaedics, 17(5):608–614.

Dickens, W. E. and Smith, M. F. (2006). Validation of a

visual gait assessment scale for children with hemi-

plegic cerebral palsy. Gait & Posture, 23(1):78–82.

Embrey, D. G., Yates, L., and Mott, D. H. (1990). Effects of

neuro-developmental treatment and orthoses on knee

ﬂexion during gait: A single-subject design. Physical

Therapy, 70(10):626–637.

Frenkel-Toledo, S., Giladi, N., Peretz, C., Herman, T., Gru-

endlinger, L., and Hausdorff, J. (2005). Effect of gait

speed on gait rhythmicity in Parkinson’s disease: vari-

ability of stride time and swing time respond differ-

ently. Journal of NeuroEngineering and Rehabilita-

tion, 2(1):23.

Gouwanda, D. and Senanayake, A. S. M. N. (2011).

Identifying gait asymmetry using gyroscopes-a cross-

correlation and normalized symmetry index approach.

Journal of Biomechanics, 44(5):972 – 978.

Hausdorff, J. and Ladin, Z. amd Wei, J. (1995). Footswitch

system for measurement of the temporal parameters

of gait. Journal of Biomechanics, 28(3):347–351.

Kawamura, C. M., Filho, M. C. M., Barreto, M. M., Asa, S.

K. P., Juliano, Y., and Novo, N. F. (2007). Comparison

between visual and three-dimensional gait analysis in

patients with spastic diplegic cerebral palsy. Gait &

Posture, 25(1):18–24.

Kempen, J., de Groot, V., Knol, D., Polman, C., Lankhorst,

G., and Beckerman, H. (2011). Community walking

can be assessed using a 10 metre timed walk test. Mul-

tiple Sclerosis Journal.

Kennedy, D., Stratford, P., Wessel, J., Gollish, J., and Pen-

ney, D. (2005). Assessing stability and change of four

performance measures: a longitudinal study evaluat-

ing outcome following total hip and knee arthroplasty.

BMC Musculoskeletal Disorders, 6(1):3.

Koman, L., Mooney, J., Smith, B., Goodman, A., and Mul-

vaney, T. (1993). Management of cerebral palsy with

botulinum-A toxin: preliminary investigation. Jour-

nal of pediatric orthopedics, 13(4):489–95.

Kristensen, M. T., Foss, N. B., and Kehlet, H. (2007).

Timed “up & go” test as a predictor of falls within

6 months after hip fracture surgery. Physical Therapy,

87(1):24–30.

Lofterød, B. and Terjesen, T. (2008). Results of treatment

when orthopaedic surgeons follow gait-analysis rec-

ommendations in children with cp. Developmental

Medicine & Child Neurology, 50(7):503–509.

Maathuis, K. G. B., van der Schans, C. P., van Iperen, A.,

Rietman, H. S., and Geertzen, J. H. B. (2005). Gait

in children with cerebral palsy: Observer reliability of

physician rating scale and edinburgh visual gait anal-

ysis interval testing scale. Journal of Pediatric Or-

thopaedics, 25(3).

Macnicol, M., McHardy, R., and Chalmers, J. (1980). Ex-

ercise testing before and after hip arthroplasty. Jour-

nal of Bone and Joint Surgery - British Volume, 62-

B(3):326–331.

Mayagoitia, R., L

otters, J., Veltink, P., and Hermens, H.

(2002a). Standing balance evaluation using a triaxial

accelerometer. Gait & Posture, 16(1):55–59.

Mayagoitia, R., Nene, A. V., and Veltink, P. H. (2002b).

Accelerometer and rate gyroscope measurement of

kinematics: an inexpensive alternative to optical mo-

tion analysis systems. Journal of Biomechanics,

35(4):537–542.

McGraw, K. O. and Wong, S. P. (1996). Forming inferences

A WEARABLE GAIT ANALYSIS SYSTEM USING INERTIAL SENSORS PART II - Evaluation in a Clinical Setting

about some intraclass correlation coefﬁcients. Psycho-

logical Methods, 1:30–46.

Mercer, J. A., Devita, P., Derrick, T. R., and Bates,

B. T. (2003). Individual effects of stride length

and frequency on shock attenuation during running.

Medicine & Science in Sports & Exercise, 35(2):307–

313.

Moe-Nilssen, R. and Helbostad, J. L. (2004). Estimation

of gait cycle characteristics by trunk accelerometry.

Journal of Biomechanics, 37(1):121 – 126.

Morais Filho, M. C., Yoshida, R., Carvalho, W. S., Stein,

H. E., and Novo, N. F. (2008). Are the recommenda-

tions from three-dimensional gait analysis associated

with better postoperative outcomes in patients with

cerebral palsy? Gait & Posture, 28(2):316–322.

Nordin, E., Lindelf, N., Rosendahl, E., Jensen, J., and

Lundin-Olsson, L. (2008). Prognostic validity of the

timed up-and-go test, a modiﬁed get-up-and-go test,

staff’s global judgement and fall history in evaluating

fall risk in residential care facilities. Age and Ageing,

37(4):442–448.

Ong, A., Hillman, S., and Robb, J. (2008). Reliability and

validity of the edinburgh visual gait score for cerebral

palsy when used by inexperienced observers. Gait &

Posture, 28(2):323–326.

Palombaro, K. M., Craik, R. L., Mangione, K. K., and Tom-

linson, J. D. (2006). Determining meaningful changes

in gait speed after hip fracture. Physical Therapy,

86(6):809–816.

Potter, J. M., Evans, A. L., and Duncan, G. (1995). Gait

speed and activities of daily living function in geriatric

patients. Archives of Physical Medicine and Rehabil-

itation, 76(11):997 – 999.

Read, H. S., Hazlewood, M. E., Hillman, S. J., Prescott,

R. J., and Robb, J. E. (2003). Edinburgh visual gait

score for use in cerebral palsy. Journal of Pediatric

Orthopaedics, 23(3):296–301.

Russell, E., Braun, B., and Hamill, J. (2010). Does stride

length inﬂuence metabolic cost and biomechanical

risk factors for knee osteoarthritis in obese women?

Clinical Biomechanics, 25(5):438–443.

Sabatini, A. M., Martelloni, C., S., S., and F., C. (2005). As-

sessment of walking features from foot inertial sens-

ing. IEEE Transactions on Biomedical Engineering,

52(3):486–494.

Salarian, A., Russmann, H., Vingerhoets, F., Dehollain, C.,

Blanc, Y., Burkhard, P., and Aminian, K. (2004). Gait

assessment in Parkinson’s disease: Toward an ambula-

tory system for long-term monitoring. IEEE Transac-

tions on Biomedical Engineering, 51(8):1434 –1443.

Saremi, K., Marehbian, J., Yan, X., Regnaux, J.-P.,

Elashoff, R., Bussel, B., and Dobkin, B. H. (2006).

Reliability and validity of bilateral thigh and foot ac-

celerometry measures of walking in healthy and hemi-

paretic subjects. Neurorehabilitation and Neural Re-

pair, 20(2):297–305.

Senden, R., Grimm, B., Heyligers, I., Savelberg, H., and

Meijer, K. (2009). Acceleration-based gait test for

healthy subjects: Reliability and reference data. Gait

& Posture, 30(2):192 – 196.

Silver, K., Macko, R., Forrester, L., Goldberg, A., and

Smith, G. (2000). Effects of aerobic treadmill train-

ing on gait velocity, cadence, and gait symmetry in

chronic hemiparetic stroke: A preliminary report.

Neurorehabilitation and Neural Repair, 14(1):65–71.

Simon, S. R. (2004). Quantiﬁcation of human motion: gait

analysis - beneﬁts and limitations to its application to

clinical problems. Journal of Biomechanics, 37:1869–

1880.

Toro, B., Nester, C., and Farren, P. (2003). A review of ob-

servational gait assessment in clinical practice. Phys-

iotherapy Theory and Practice, 19(3):137–149.

Wren, T. A. L., Rethlefsen, S. A., Healy, B. S., Do, K. P.,

Dennis, S. W., and Kay, R. M. (2005). Reliability and

validity of visual assessments of gait using a modi-

ﬁed physician rating scale for crouch and foot contact.

Journal of Pediatric Orthopaedics, 25(5):646–650.

BIOSIGNALS 2012 - International Conference on Bio-inspired Systems and Signal Processing