//Logo Image
Author: Rung-Hung Su, Yeh-Liang Hsu, Lung Chan, Hanjun Lin, Che-Chang Yang (2014-09-09); recommended: Yeh-Liang Hsu (2014-09-09).
Note: This paper is published in Biomedical Engineering: Applications, Basis and Communications, 26(2), DOI: 10.4015/S1016237214500318

Assessing abnormal gaits of Parkinson’s disease patients using a wearable motion detector

ABSTRACT

Accelerometers have been widely used in wearable systems for gait analysis. Several gait cycle parameters are provided to quantify the level of gait regularity and symmetry. This study attempts to assess abnormal gaits of Parkinson disease (PD) patients based on the gait cycle parameters derived in real-time from an accelerometry-based wearable motion detector. The results of an experiment with 25 healthy young adults showed that there were significant differences between gait cycle parameters of normal gaits and abnormal gaits derived from the wearable motion detector. Five PD patients diagnosed as Hoehn & Yahr stage I to II were recruited. It is difficult to collect data of abnormal gaits of the PD patients; therefore, ranges of the gait cycle parameters of abnormal gaits of PD patients were estimated statistically based on the “lower confidence limit” of the gait cycle parameters of their normal gaits. These results may lead to the future development of wearable sensors enabling real-time recognition of abnormal gaits of PD patients. Ambulatory rehabilitation, gait assessment and personal telecare for people with gait disorders are also possible applications.

Keywords: gait cycle parameters; Parkinsons disease; wearable accelerometry system

INTRODUCTION

Individuals with degenerative mobility, e.g., Parkinson’s disease (PD) patients or older adults usually have gait disorders such as reduced walking speeds with increased cadences, reduced step/stride lengths, and increased inter-stride variability1. PD is a progressive neurological condition characterized by hypokinesia (reduced movement), akinesia (absent movement), tremors, rigidity and postural instability. These movement disorders are associated with a slow short-stepped, shuffling gait pattern2. PD patients in the advanced stage may encounter episodic gait disturbances, like festinating or even freezing of gaits which can lead to falling and adverse health outcomes3, 4.

Falling is a frequent complication for patients with PD. Falling will lead to an incapacitating fear of falling again, because the patient fears additional complications such as severe fractures. Therefore, it is important to develop a system designed for preventing PD patients from falling. In order to achieve this aim, the normal and abnormal gaits of PD patients should be defined and distinguished. If a PD patient is walking normally and then develop an abnormal gait, a warning signal can be issued in real-time to remind the patient to adjust his or her stride. For this purpose, gait cycle parameters, which can quantify and describe the characteristic of each gait, need to be defined first. In related studies, cadence, regularity, rhythm and symmetry are important gait cycle parameters which can apparently be altered in walking patterns among patients with varied mobility3-5.

Gait monitoring and analysis techniques for measuring the above gait cycle parameters have been widely developed and studied. For example, gait dynamics can be accurately measured using optical motion capture systems which use high-speed infrared cameras to record the three-dimensional positions of reflective markers attached to the joints and segments of the human body6. Gait detection techniques using pressure sensors embedded in a walkway have also been used7. These techniques can detect foot contact (heel strike and toe-off) and evenness of foot pressure distribution in an effort to investigate temporal gait parameters. However, these systems are expensive and require sophisticated instrumentation and specialized personnel. So, the uses of such systems are usually limited to laboratory or clinical environments. Simpler systems based on pressure detection, such as a portable in-shoe pressure measurement system have also been introduced6, 8. The systems using in-shoe pressure detection can only provide simple temporal gait measures while video-based systems can provide temporal and spatial gait measures, even including accurate measurement of the motion of lower limbs and body.

In recent years, accelerometers have been widely accepted as useful and practical sensors used as wearable devices to measure gait. For example, Yang et al.9 provided a cost-effective approach to real-time gait monitoring. They developed a wearable accelerometry system for real-time gait cycle parameters recognition. A waist-mounted wearable motion detector (WMD) was designed to measure trunk accelerations during walking. The autocorrelation procedure is implemented in the WMD for online calculating the real-time gait cycle parameters, including cadence, step regularity, stride regularly and step symmetry in real-time. These parameters can be sent out via Zigbee transmission for further recording, processing and/or reaction.

Though the gait monitoring and analysis techniques for PD patients have been well studied and described, to the best of our knowledge, the recognition of normal and abnormal gaits for PD patients from their gait cycle parameters has not been thoroughly investigated. The aim of this paper is to access the abnormal gaits of PD patients based on the gait cycle parameters measured by the WMD. First an experiment with 25 healthy young adults was designed to compare the gait cycle parameters of normal and abnormal gaits derived by the WMD. In this experiment, abnormal gaits were generated by constraining the movement of one leg of the testers. The results showed that there were significant differences between gait cycle parameters of normal and abnormal gaits.

Five PD patients diagnosed as Hoehn & Yahr stage I to II were recruited to collect data related to their gait cycle parameters. It is difficult to collect data of abnormal gaits of the PD patients. Therefore, ranges of the gait cycle parameters of normal and abnormal gaits of PD patients were estimated statistically based on the “lower confidence limit” of the gait cycle parameters of their normal gaits.

In this paper, the next section briefly introduces the structure of the WMD. Section 3 describes the experiment using young healthy adults to compare gait cycle parameters of normal and abnormal gaits. Section 4 describes and experiment to estimate the range of abnormal gaits based on the collected normal gait data from PD patients. Section 5 concludes the paper and discusses possible future research.

RESEARCH METHOD

Wearable Motion Detector

The WMD developed by Yang et al.9 is a single waist-mounted device which measures trunk accelerations during walking. The best position for wearing the device is near the mid-body between the anterior superior iliac spine and the right iliac crest around the pant beltline. Powered by 3 AAA batteries (DC 4.5V), the WMD measures 90 mm × 50 mm × 25 mm and weighs 120 g. Fig. 1 shows the WMD prototype, which has the following three modules:

Tri-axial accelerometer (KXPA4-2050, Kionix): senses accelerations with a sensitivity of 660 mV/g over the selected range of ±2 g. The output of the accelerometer module is first low-pass filtered at the cut-off frequency of 50 Hz to reduce signal noise.

PIC microcontroller (PIC18LF6722, Microchip): samples the analog output signals via a 10-bit A/D conversion at a sampling rate of 50 Hz. Real-time signal processing can be done with the PIC microcontroller.

Wireless ZigBee RF (XBee 2.0, Digi International): enables wireless data transmission to a PC via a 2.4 GHz ZigBee protocol.

Fig. 1 The prototype of the WMD

The measured accelerations during walking reveal periodic signal patterns. Autocorrelation procedure has been widely used to estimate the repeating characteristics over a signal sequence containing periodic patterns and irregular noise for gait cycle parameters analysis10-12.

Consider a time-discrete acceleration sequence containing N signal points [ ], Equation (1) calculates the autocorrelation coefficient , which is the sum of the products of  multiplied by another signal  at the given phase shift m. The phase shift m can be either positive or negative integers from 0 to N-1, or from 0 to 1-N. Therefore, from an N-point acceleration sequence, its autocorrelation sequence

can be represented by 2N-1 autocorrelation coefficients obtained at every phase shift m. The autocorrelation sequence can either be “biased” or “unbiased”. The unbiased autocorrelation sequence as shown in Equation (2) is preferred because the biased method generates noticeable attenuation of coefficient values next to the zero phase shift from a limited number of data.

                                                                                                        (1)

                                                                                    (2)

The segments from  to  and from  to  in an autocorrelation sequence are symmetric with its zero phase shift  located at the center of the sequence. Normalized to 1 at the zero phase shift , only the right half segment  to  of the autocorrelation sequence is considered for simplicity.

Fig. 2 depicts an example of an autocorrelation sequence computed from the vertical accelerations measured at waist during normal walking paces. A signal sequence with perfectly repetitive pattern produces its autocorrelation sequence containing the peak magnitudes identical to its zero phase shift at every dominant period. The first coefficient peak D1 next to the zero phase shift indicates the first dominant period, and the second peak D2 indicates the second dominant period. The peaks D1 and D2 can be detected by a simple derivative-based method and zero-crossing identification, which are commonly used in detecting peaks in physiologic signals, such as PQRST points in the Electrocardiogram (ECG) signals.

Fig. 2 An autocorrelation sequence computed from the vertical acceleration during walking

The autocorrelation procedure is implemented in the WMD for online calculating the real-time gait cycle parameters. Four gait cycle parameters are derived from the measured trunk accelerations:

Cadence (C): defined as the number of steps taken per minute. Let L be the number of steps taken over the time period t (in second). Cadence can thus be expressed as Equation (3). The number of steps L can be expressed as the number of the total samples N divided by the number of the coefficients n between the zero phase shift and the first dominant period, i.e., . The time period t during walking can also be alternatively expressed as N divided by the sampling frequency f, i.e., . As a result, C can be estimated by Equation (4), which was given by Moe-Nilssen et al.10

                                                                                                                (3)

                                                                                        (4)

Step regularity (D1): The regularity of each step. According to Moe-Nilssen et al.10, the magnitude D1 can be represented the step regularity. This is because the first dominant period indicates the maximal similarity between the acceleration sequence and its m-point shifted duplicate. The m-point span approximates the duration of a step. The value of D1 ranges from 0 to 1 and should be close to 1 in a normal gait pattern.

Stride regularity (D2): defined as the regularity of two steps. The second dominant period indicates the maximal similarity between the acceleration sequence and its 2-step shifted sequence, and therefore the magnitude D2 can also be represented the stride regularity. Note that the first and second dominant periods do not represent which of the steps (left-leg or right-leg) as there is no such information given in the autocorrelation procedure. The value of D2 ranges from 0 to 1 and should be close to 1 in a normal gait pattern.

Step symmetry (S): defined as the symmetry between two steps of both legs. The calculation of step symmetry is  if , and  when . Note that this definition is slightly altered from the definition originally given by Moe-Nilssen et al.10 and its modified version by Yang et al.11 The value of S ranges from 0 to 1 and should be close to 1 in a normal gait pattern.

RESULTS AND ANALYSIS

Comparing Gait Cycle Parameters of Normal and Abnormal Gaits

This section describes an experiment with healthy young adults to compare the difference between the gait cycle parameters of normal and abnormal gaits measured by the WMD. Twenty-five healthy young adults (13 males and 12 females, 22 ± 3 yrs) were recruited for gait data collection. A 40-meters-walk test was conducted in a laboratory for each tester. In this test, the testers wore the WMD at their waists while walking on a level walkway using normal and abnormal gaits. Note that the abnormal gaits were generated by constraining the movement of one leg with an aggravating legging (Alex C-21, weights 4.54 kg) and an adjustable knee pad (Alex T-09). Three-dimensional accelerations, vertical (VT), antero-posterior (AP) and medio-lateral (ML), were recorded at the sampling rate of 50Hz. The initiation of data collection by the WMD triggered the start of a synchronized video recording during the test for gait observation and cadence validation.

Table 1 and 2 show the gait cycle parameters of normal and abnormal gaits for each tester, where Ca, D1a, D2a, and Sa are the gait cycle parameters in normal gaits; Cb, D1b, D2b, and Sb are the gait cycle parameters of abnormal gaits; , ,  and  are the difference in gait cycle parameters between normal and abnormal gaits. From Table 1 and 2, the mean value of cadence (C) dropped 3.72% from normal to abnormal gaits, the mean value of step regularity (D1) dropped 23.70%, the mean value of stride regularity (D2) dropped 1.58%, and the mean value of step symmetry (S) dropped 23.04%. As expected, the testers had lower cadence, step regularity, stride regularity and step symmetry when the movement of one leg was constrained.

Table 1. Cadence (C) of normal and abnormal gaits

Tester

Gait cycle parameter

Tester

Gait cycle parameter

Ca

Cb

Ca

Cb

1

105.900

104.470

1.430

15

111.800

110.440

1.360

2

114.950

93.800

21.150

16

117.600

116.700

0.900

3

107.470

101.230

6.240

17

115.790

105.990

9.800

4

105.370

106.600

-1.230

18

121.150

114.950

6.200

5

113.700

109.910

3.790

19

106.600

105.960

0.640

6

120.450

116.900

3.550

20

113.830

107.700

6.130

7

110.150

105.030

5.120

21

119.700

119.350

0.350

8

113.060

104.380

8.680

22

109.390

104.460

4.930

9

106.480

107.070

-0.590

23

111.810

110.400

1.410

10

113.000

108.820

4.180

24

119.700

117.500

2.200

11

111.060

108.900

2.160

25

113.000

113.000

0.000

12

115.600

104.070

11.530

 

 

 

 

13

116.250

110.430

5.820

Mean

112.788

108.589

4.200

14

105.900

106.660

-0.760

Std. dev.

4.842

5.716

4.889

Table 2. Step regularity (D1), stride regularity (D2) and step symmetry (S) of normal and abnormal gaits

Tester

Gait cycle parameters

D1a

D1b

D2a

D2b

Sa

Sb

1

0.851

0.540

0.312

0.909

0.900

0.009

0.930

0.610

0.320

2

0.674

0.301

0.373

0.845

0.767

0.078

0.800

0.400

0.400

3

0.643

0.565

0.079

0.917

0.898

0.019

0.700

0.590

0.120

4

0.847

0.583

0.244

0.912

0.883

0.028

0.910

0.640

0.270

5

0.858

0.740

0.117

0.878

0.886

-0.008

0.970

0.840

0.130

6

0.754

0.500

0.254

0.873

0.862

0.011

0.870

0.580

0.290

7

0.650

0.522

0.128

0.928

0.842

0.086

0.700

0.620

0.080

8

0.606

0.498

0.109

0.792

0.869

0.077

0.780

0.570

0.210

9

0.839

0.579

0.260

0.879

0.902

-0.023

0.870

0.640

0.310

10

0.853

0.777

0.076

0.883

0.895

-0.012

0.950

0.870

0.220

11

0.642

0.568

0.074

0.912

0.905

0.007

0.700

0.630

0.070

12

0.676

0.574

0.101

0.849

0.873

-0.024

0.810

0.660

0.150

13

0.888

0.802

0.086

0.899

0.897

0.002

0.980

0.850

0.130

14

0.880

0.625

0.255

0.925

0.909

0.017

0.950

0.690

0.260

15

0.861

0.783

0.078

0.910

0.909

0.000

0.950

0.860

0.090

16

0.675

0.421

0.255

0.878

0.889

-0.011

0.770

0.480

0.290

17

0.670

0.492

0.178

0.811

0.778

0.034

0.830

0.560

0.270

18

0.770

0.513

0.258

0.893

0.848

0.044

0.860

0.610

0.250

19

0.879

0.636

0.243

0.916

0.934

-0.017

0.940

0.680

0.260

20

0.611

0.578

0.032

0.942

0.876

0.066

0.650

0.660

-0.010

21

0.707

0.474

0.233

0.845

0.911

-0.066

0.840

0.520

0.320

22

0.700

0.390

0.310

0.842

0.741

0.101

0.830

0.500

0.330

23

0.643

0.582

0.061

0.929

0.845

0.085

0.700

0.700

0.000

24

0.731

0.462

0.269

0.881

0.894

-0.013

0.830

0.520

0.310

25

0.857

0.822

0.034

0.909

0.881

0.028

0.940

0.920

0.020

Mean

0.751

0.573

0.177

0.886

0.872

0.021

0.842

0.648

0.204

Std. dev.

0.100

0.131

0.101

0.038

0.047

0.042

0.099

0.133

0.116

To estimate the decreasing value of gait cycle parameters from normal to abnormal gaits, the following hypotheses testing were implemented:

 versus  i = 1, 2, 3, 4

where , ,  and are the average of the decreasing value of C, D1, D2 and S, respectively;  and  are each designated decreasing values.

From the sample means and sample standard deviations (Table 1 and 2), the p-value for the tests with  are calculated under the assumption the gait cycle parameters are normally distributed and the obtained values were , ,  and , respectively. If the level of Type I error  (i.e., the chance of incorrectly judging the null hypothesis as alternative hypothesis) is 0.01, the conclusions for all tests reject null hypotheses ( ). This implies that, at a 99% confidence level, the gait cycle parameters of normal and abnormal gaits measured by the WMD are significantly different. Table 3 shows the maximum decreasing values of the four gait cycle parameters when Type I error 0.1, 0.05, 0.025 and 0.01. For example, at a 95% confidence level ( 0.05), d1 (drop in cadence) is 2.527, d2 (drop in step regularity) is 0.142, d3 (drop in stride regularity) is 0.006, and d4 (drop in symmetry) is 0.164.

Table 3. The maximum decreasing values for four gait cycle parameters
when 0.1, 0.05, 0.025 and 0.01.

Type I error

Confidence level
100 × (1– )%

Decreasing values

d1

d2

d3

d4

0.1

90%

2.911

0.150

0.010

0.173

0.05

95%

2.527

0.142

0.006

0.164

0.025

97.5%

2.181

0.135

0.003

0.156

0.01

99%

1.726

0.126

0.000

0.146

Estimating Gait Cycle Parameters of Abnormal Gaits of PD Patients

After confirming that there were significant differences between gait cycle parameters of normal gaits and abnormal gaits, a gait data-collection experiment with PD patients was designed to assess abnormal gaits of PD patients based on the gait cycle parameters derived by the WMD.

Five elderly PD patients (four males and one female, 65-72 yrs.) diagnosed as H&Y stage I-II with self-mobility were recruited for data collection of their normal gaits. This experiment was approved by the Institutional Review Board (IRB) at the Far Eastern Memorial Hospital (FEMH-IRB-100076-F) and was performed in accordance with the ethical standards of the 1964 Declaration of Helsinki. Participants gave informed consent prior to their inclusion in the study and were provided with required information about the experiment.

In this experiment, a 10-meters-walk test was conducted in a rehabilitation center for each tester. Table 4 shows step regularity (D1), stride regularity (D2) and step symmetry (S) of normal gaits for five samples ( ) of the five ( ) testers. Note the cadence data was not recorded because of the large individual difference of each tester in cadence. Comparing Table 4 with Table 1 and 2, step regularity, stride regularity of PD patients are lower than those of healthy young subjects, while the mean value of step symmetry of PD patients is similar to that of healthy young subjects.

Table 4. The data of normal gaits of D1, D2 and S for five PD patients

Patient

Order

Gait cycle parameters

Patient

Order

Gait cycle parameters

i

j

D1

D2

S

i

j

D1

D2

S

1

1

0.6044

0.6091

0.9924

4

1

0.5705

0.5928

0.9625

2

0.5383

0.5273

0.9796

2

0.5468

0.5444

0.9956

3

0.6929

0.6516

0.9405

3

0.1943

0.3812

0.5098

4

0.6321

0.6653

0.9502

4

0.6268

0.6567

0.9534

5

0.5431

0.6340

0.8567

5

0.5145

0.5005

0.9728

2

1

0.5012

0.6603

0.7591

5

1

0.4507

0.3547

0.7869

2

0.4084

0.4398

0.9286

2

0.5312

0.5481

0.9692

3

0.4328

0.2995

0.6920

3

0.5533

0.5465

0.9877

4

0.4895

0.4848

0.9903

4

0.4946

0.5486

0.9015

5

0.5126

0.6340

0.8085

5

0.5330

0.4615

0.8659

3

1

0.5221

0.5973

0.8741

 

 

 

 

 

2

0.4938

0.6025

0.8195

 

 

 

 

 

3

0.5180

0.4569

0.8821

 

 

 

 

 

4

0.5633

0.6123

0.6200

Mean

0.5077

0.5342

0.8659

5

0.2233

0.3447

0.6477

Std. dev.

0.1092

0.1074

0.1325

It is difficult to collect data of abnormal gaits of PD patients because of the danger of falls. Therefore, in this experiment, criteria for distinguishing between normal and abnormal gaits of PD patients were estimated statistically based on the “lower confidence limit” of the gait cycle parameters of their normal gaits. Note that the conservative lower confidence limit can be regarded as lower bound of normal gait. From the data related to normal gaits in Table 4, Table 5 shows 90%, 95%, 97.5% and 99% lower confidence limits calculated for the average of D1, D2 and S (see Appendix for details of calculating the lower confidence limits). For example, the lower confidence limit of D1 is 0.4797 when the required confidence level is set at 90%, the ranges of normal and abnormal gaits for step regularity are  and , respectively. Figs. 3-5 depict the ranges of normal and abnormal gaits for D1, D2 and S. The raw data of five PD patients are also plotted into Figures 3-5 so that it is better observed on the relationship between normal and abnormal data. Note that some raw data fall in abnormal area due to the large variation of data, especially D2 and S.

Table 5. The lower confidence limits of gait cycle parameters of normal gaits of PD patients

Type I error

Confidence level

100 × (1– )%

Gait cycle parameters

D1

D2

S

0.1

90%

0.4797

0.5059

0.8317

0.05

95%

0.4712

0.4974

0.8214

0.025

97.5%

0.4636

0.4897

0.8121

0.01

99%

0.4543

0.4802

0.8007

Fig. 3 Ranges of normal and abnormal gaits for D1

Fig. 4 Ranges of normal and abnormal gaits for D2

Fig. 5 Ranges of normal and abnormal gaits for S

CONCLUSIONS

This study attempts to assess abnormal gaits of PD patients based on the gait cycle parameters derived in real-time from a single wearable motion detector. The results of the first experiment with 25 healthy young adults in this paper showed that, there were significant differences between gait cycle parameters of normal gaits and abnormal gaits derived from the wearable motion detector. The results of the second experiment with five PD patients provided the lower confidence limits for the average of step regularity, stride regularity and symmetry at different confidence levels, which can be used the derive the ranges of the gait cycle parameters of abnormal gaits.

These results may lead to the future development of wearable sensors enabling real-time recognition of abnormal gaits of PD patients. The autocorrelation procedure is implemented in the WMD for online calculating the real-time gait cycle parameters. If a PD patient is walking normally and then starts to develop an abnormal gait, the statistically estimated ranges of all three gait cycle parameters at different confidence levels can be used for the judgment. A warning signal can be issued in real-time to remind the patient to adjust his or her stride if continuous gait cycle parameters in the abnormal range are detected.

In future work, more data will be accumulated to fine tune the thresholds of abnormal gait cycle parameters, so that the WMD can be a practical sensor for ambulatory rehabilitation, gait assessment and personal telecare for PD patients a d people with gait disorders.

ACKNOWLEDGMENT

We gratefully acknowledge the support of this research by the Division of Neurology and Orthopedics in Far Eastern Memorial Hospital, Taiwan.

APPENDICES

Lower Confidence Limit

A “confidence interval” gives an estimated range of values which is likely to include the underlying population parameter if independent samples are taken repeatedly from the same population, with the specified level of probability. The “lower confidence limit” is the lower end of a confidence interval. Under the assumption that the observations are normally distributed, if a multiple samples of m groups each of size n is given as , where i = 1, 2, …, m and j = 1, 2, …, n, the  lower confidence limit for mean of observation is

,

where , , ,  and  is the upper  quantile of a t distribution with m(n-1) degrees of freedom satisfying .

REFERENCES

1.        Lowry KA, Smiley-Oyen AL, Carrel AJ, Kerr JP, Walking stability using harmonic ration in Parkinson’s disease, Mov Disord  24(2):261-267, 2009.

2.        Morris ME, Huxham F, McGinley J, Dodd K, Iansek R, The biomechanics and motor control of gait in Parkinson disease, Clin Biomech 16(6):459-470, 2001.

3.        Hausdorff JM, Gait dynamics, fractals and falls: Finding meaning in the stride-to-stride fluctuations of human walking, Hum Mov Sci 26(4):555-589, 2007.

4.        Hausdorff JM, Gait dynamics in Parkinson’s disease: Common and distinct behavior among stride length, gait variability, and fractal-like scaling, Chaos 19(2):026113. doi: 10.1063/1.3147408, 2009.

5.        Plotnik M, Hausdorff JM, The role of gait rhythmicity and bilateral coordination of stepping in the pathophysiology of freezing of gait in Parkinson’s disease, Mov Disord  23(Suppl 2):S444-450, 2008.

6.        Femery V, Moretto P, Renaut H, Thevenon A, Lensel G, Measurement of plantar pressure distribution in hemiplegic children: changes to adaptive gait patterns in accordance with deficiency, Clin Biomech  17(5):406-413, 2002.

7.        Menz HB, Latt MD, Tiedemann A, Kwan MMS, Lord SR, Reliability of the GAITRite walkway system for the quantification of temporal-spatial parameters of gait in young and older people, Gait Posture 20(1):20-25, 2004.

8.        Hausdorff JM, Ladin Z, Wei JY, Footswitch system for measurement of the temporal parameters of gait, J Biomech 28(3):347-351, 1995.

9.        Yang CC, Hsu YL, Shih KH, Lu JM, Real-Time Gait Cycle Parameters Recognition Using a Wearable accelerometry system, Sensors 11(8):7314-7326, 2011.

10.     Moe-Nilssen R, Helbostad JL, Estimation of gait cycle characteristics by trunk accelerometry, J Biomech 37(1):121-126, 2004.

11.     Yang M, Zheng H, Wang H, McClean S, Hall J, Harris N, Assessing accelerometer based gait features to support gait analysis for people with complex regional pain syndrome, In Proceedings of the 3rd International Conference on PErvasive Technologies Related to Assistive Environments, Samos, Greece, 13-15 June 2010; ACM: New York, USA, 2010; No. 48.

12.     Keenan DB, Wilhelm FH, Classification of locomotor activity by acceleration measurement: validation in Parkinson disease, Biomed Sci Instrum 41:329-334, 2005.