A Comparison of Algorithms for Body-Worn Sensor-Based ...

Journal of Applied Biomechanics, 2012, 28, 349-355 © 2012 Human Kinetics, Inc.

A Comparison of Algorithms for Body-Worn Sensor-Based Spatiotemporal Gait Parameters to the GAITRite Electronic Walkway Barry R. Greene,1 Timothy G. Foran,2 Denise McGrath,3 Emer P. Doheny,1 Adrian Burns,1 and Brian Caulfield3,4 1Intel

Laboratories, Co. Kildare, and the TRIL Centre; 2St. James’s Hospital, Dublin; College Dublin; 4CLARITY: Centre for Sensor Web Technologies

3University

This study compares the performance of algorithms for body-worn sensors used with a spatiotemporal gait analysis platform to the GAITRite electronic walkway. The mean error in detection time (true error) for heel strike and toe-off was 33.9 ± 10.4 ms and 3.8 ± 28.7 ms, respectively. The ICC for temporal parameters step, stride, swing and stance time was found to be greater than 0.84, indicating good agreement. Similarly, for spatial gait parameters—stride length and velocity—the ICC was found to be greater than 0.88. Results show good to excellent concurrent validity in spatiotemporal gait parameters, at three different walking speeds (best agreement observed at normal walking speed). The reported algorithms for body-worn sensors are comparable to the GAITRite electronic walkway for measurement of spatiotemporal gait parameters in healthy subjects. Keywords: body-worn sensors, GAITRite, SHIMMER, gait analysis Quantitative measurement of gait is a valuable tool in clinical practice. Many studies have demonstrated that it can be used to discriminate presence of specific gait deviations associated with injury and disease, predict risk of falling, and quantify improvements due to rehabilitation (Delahunt et al., 2006; Paul et al., 2007; Shrader et al., 2009). However, gait analysis has traditionally involved the use of expensive laboratory-based motion capture systems that require skilled personnel for implementation and data processing. Recent years have seen the introduction of less expensive systems, such as GAITRite (CIR Systems, Inc.), which is a capacitive force-sensitive electronic walkway that facilitates quantitative analysis of spatiotemporal gait variables using a pressure-sensitive walkway. These systems do not yield the rich kinematic data provided by marker-based motion capture systems linked to force platforms, yet they still provide valuable Barry R. Greene is with Applied Technology & Design, Intel Labs, Co. Kildare, Ireland, and with the TRIL Centre. Timothy G. Foran is with the Department of Medical Physics and Clinical Engineering, St. James’s Hospital, Dublin, Ireland. Denise McGrath is with the School of Physiotherapy and Performance Science, University College Dublin, Ireland. Emer P. Doheny and Adrian Burns are with Applied Technology & Design, Intel Labs, Co. Kildare, Ireland, and with the TRIL Centre. Brian Caulfield is with the CLARITY: Centre for Sensor Web Technologies and the School of Physiotherapy and Performance Science, University College Dublin, Ireland.

clinical data (Kressig et al., 2006). Furthermore, they represent a significant advance over manual methods of quantitative spatiotemporal gait analysis (Youdas et al., 2010), do not require body-worn markers, and have been validated against gold standard methods, such as laboratory-grade force plates and optical motion capture systems for quantitative spatiotemporal gait analysis (Bilney et al., 2003; Webster et al., 2005). Therefore, they have achieved widespread clinical use. Recent trends in field monitoring, particularly in the cases of sports science and in care of the elderly, have presented a need for low-cost, portable gait analysis technology. A major limitation of GAITRite is its relatively high cost and location in a clinical laboratory environment. Body-worn inertial sensors consisting of either accelerometers (Zijlstra et al., 2003), gyroscopes (Aminian et al., 2002; Najafi et al., 2002; Sabatini et al., 2005; Tong et al., 1999), or a combination of both (Jasiewicz et al., 2006) may be suitable for portable, low-cost spatiotemporal gait analysis. Recently, two algorithms for the analysis of body-worn sensor data have been reported. The first is a method for estimating temporal gait parameters from shank-mounted gyroscopes that can adapt to varying walking speeds and gyroscope signal quality (Greene et al., 2010a, 2010b). This method was found to compare well to a force plate and an optical motion capture system in healthy participants. Results show that the mean true error between the adaptive gyroscope algorithm and force plate was –4.5 ± 14.4 ms and 43.4 ± 6.0 ms for IC and TC points respectively, in healthy subjects. Similarly, the

349

350  Greene et al.

mean true error when data from the polio patient were compared against the force plate was –75.61 ± 27.53 ms and 99.20 ± 46.00 ms for IC and TC points, respectively. A comparison of temporal gait parameters derived from an optical motion analysis system showed good agreement for nine healthy subjects at four speeds. The second algorithm is a method for robustly deriving stride length and stride velocity from a single shank-mounted gyroscope (Doheny et al., 2010). Results for this algorithm compared with GAITRite showed that the two systems compared favorably, with a mean error in stride length of 0.09 ± 0.07 m, and a mean error in stride velocity of 0.11 ± 0.10 m/s. The aim of this study was to compare the performance of two novel body-worn sensor algorithms to the GAITRite electronic walkway for measurement of spatiotemporal gait parameters at a variety of self-selected walking speeds in healthy subjects. As both GAITRite and body-worn inertial sensor algorithms derive spatiotemporal parameters using different methodologies, it is important to compare concurrent results to show reliability of novel algorithms against a clinically acceptable standard. It was hypothesized that data recorded simultaneously from both instruments would demonstrate good concurrent validity (levels of agreement, ICC > 0.8) in the measurement of spatiotemporal parameters.

Method Experimental Setup The gaits of seven normal healthy subjects (3 male; age: 23–49 years, height: 161–190 cm; weight: 55–91 kg) were measured simultaneously using two gait measurement technologies: the body-worn triaxial gyroscopes and the GAITRite electronic walkway (CIR Systems Inc). Each subject gave their informed consent to take part in the study. Ethical approval was obtained. All speeds were self-selected and not controlled so as to minimize the effect of the experimental protocol on the subjects’ gait. Data were recorded while each subject performed four walks at three different self-selected speeds: normal, fast and slow, along the GAITRite electronic walkway (per subject, 12 walking trials in all). In all, 84 walking trials were completed, yielding 503 separate strides for subsequent analysis.

Body-Worn Sensor Data Acquisition Kinematic data were acquired using gyroscopes (SHIMMER; Burns et al., 2010; Shimmer Research, Dublin, Ireland) attached to the anterior shank of the left and right leg. Each sensor was positioned to capture movement about the anatomical mediolateral axis of the shank of the leg, and secured using a compression stocking, halfway on the imaginary line between the tibial tuberosity and the lateral malleolus. The mediolateral angular velocity derived from a triaxial gyroscope, sampled at 102.4 Hz, was used in this study for the detection of all gait events.

Data from each of the gyroscopes were acquired using BioMOBIUS (McGrath et al., 2009) and stored on a PC. All signal processing and data analysis was carried out off-line using the MATLAB environment (version 7.8, Natick, MA, USA). A standard calibration procedure (Ferraris et al., 1995) was used to calibrate all gyroscopes used in the study. Before further processing, the raw gyroscope signals were low pass filtered with zero-phase fifth-order Butterworth filter with a 50.2 Hz corner frequency. The acquisition system and the GAITRite electronic walkway were synchronized using a dedicated trigger output from the GAITRite system. This trigger provided a 5 V pulse on an output pin when recording on the GAITRite and was activated at the initiation and deactivated at the conclusion of each walk. The trigger was connected to the analog-to-digital input of a dedicated synchronization sensor and transmitted wirelessly to the PC. Synchronization and kinematic data were simultaneously recorded within BioMOBIUS.

Electronic Walkway Data Acquisition An electronic walkway, 5.24 m long and 0.88 m wide (GAITRite), was used in the current study to measure spatial and temporal parameters of gait. The active area of the walkway is approximately 4.27 m long and 0.61 m wide. Pressure sensors are embedded into the carpet in a grid, with 12.7 mm between the sensors. As the participant walks over the carpet, the sensors close under pressure, enabling collection of temporal and spatial footfall events. Data were sampled from the electronic walkway at a frequency of 80 Hz, allowing a temporal resolution of 12.5ms. The walkway was connected to a laptop by a serial interface cable, where the spatial and temporal characteristics of gait are processed and stored using GAITRite analysis software. The GAITRite system has itself been validated in numerous studies and has been shown to have good test–retest reliability (Bilney et al., 2003; Webster et al., 2005).

Gait Parameters Spatiotemporal parameters of gait from the sensor data were derived using two previously reported algorithms: an algorithm (Greene et al., 2010a) that adaptively applies thresholds to determine heel-strike and toe-off events from the mediolateral angular velocity of the shank and an algorithm that accurately estimates stride length and stride velocity from a single shank-mounted gyroscope (Doheny et al., 2010). A sample of the mediolateral angular velocity from a subject’s shank during slow walking is illustrated in Figure 1. The heel strike (HS) and toe-off (TO) points for each gait cycle are marked. The heel strike and toe-off characteristic points derived using the gyroscope and GAITRite systems were used to calculate stride time (s), stance time (s), swing

Body-Worn Sensors and the GAITRite Walkway   351

GAITRite as a reference, using intraclass correlation coefficient (ICC(2, k) (Shrout et al., 1979), as reported previously (Bilney et al., 2003; Hartmann et al., 2009). The mean true error and mean percentage error between the two instruments were also examined on a step-bystep basis and used to further highlight the differences between instruments (Desailly et al., 2009). The true error was defined as the difference between the HS or TO time detected using the GAITRite system and the same point detected using the body-worn sensor data. The true error for all spatiotemporal gait parameters was calculated as the difference between the body-worn sensor and GAITRite values.

Results

Figure 1 — (Top panel) Sample of mediolateral angular velocity signal derived from shank-mounted gyroscope during slow walking. (Bottom panel) Sample of mediolateral angular velocity signal derived from shank mounted gyroscope sensor during normal walking. Heel strike is identified as the first negative peak after midswing point.

time (s) and step time (s). The HS and TO points were also used, along with each subject’s height to calculate stride length (cm) and stride velocity (cm/s).

Statistical Analysis A measure of concurrent validity can be established when simultaneous measurements are made by the instrument to be validated and a previously validated tool that acts as the criterion. Levels of agreement are then calculated between both instruments. Gait parameters derived from the left and right shank gyroscopes were compared on a step-by-step basis against the data derived from the

The mean (± SD) values of gait velocity obtained from GAITRite for each self-selected walking speed are as follows: slow: 87.85 ± 15.84 cm/s, normal: 139.33 ± 18.47 cm/s, fast: 202.31 ± 19.43 cm/s. A one-way ANOVA was performed to verify that all self-selected walking speeds were significantly mutually different (F = 23,005, p = 2.8235e-009). Table 1 shows the mean (± SD) value of each spatiotemporal gait parameter for both instruments. Table 2 provides summary results in terms of true error, percentage error and ICC. The performance, in terms of true error, of the body-worn sensor algorithms compared with the GAITRite in detecting heel strike and toe-off points is examined. The mean true error in heel strike and toe-off points at normal walking speed was 28.95 ± 6.98 ms and 5.90 ± 29.67 ms, respectively. Similarly, the true error in heel strike and toe off for fast and slow walking speeds was 31.36 ± 7.85 ms, 0.63 ± 27.98 ms, 41.38 ± 13.37 ms and 4.87 ± 26.64 ms, respectively, results are depicted graphically in Figure 2. The concurrent validity as measured by ICCs for temporal parameters step, stride, swing and stance time was found to be greater than 0.84, indicating good agreement. Similarly, for the spatial gait parameters, stride length and velocity, the ICC was found to be greater than 0.88. The mean value for each performance metric for each of the gait parameters is given for each self-selected walking speed.

Table 1  Mean values taken across all subjects for each spatiotemporal gait parameter for both instruments (GAITRite and body-worn sensor algorithms) GAITRite

Body-Worn Sensor

Slow

Normal

Fast

Slow

Normal

Fast

Stride Time (s)

1.42 ± 0.20

1.07 ± 0.08

0.90 ± 0.06

1.46 ± 0.19

1.12 ± 0.08

0.94 ± 0.05

Stance Time (s)

0.90 ± 0.13

0.66 ± 0.06

0.54 ± 0.04

0.92 ± 0.14

0.68 ± 0.06

0.54 ± 0.03

Swing Time (s)

0.52 ± 0.07

0.41 ± 0.03

0.37 ± 0.02

0.54 ± 0.06

0.44 ± 0.04

0.40 ± 0.03

0.57 ± 0.04

0.48 ± 0.02

Step Time (s)

0.71 ± 0.10

0.54 ± 0.04

0.45 ± 0.03

0.74 ± 0.09

Stride Length (cm)

121.48 ± 10.61

148.45 ± 12.64

182.20 ± 19.04

129.30 ± 9.91

152.76 ± 10.73 163.44 ± 10.92

Stride Velocity (cm/s)

87.85 ± 15.84

139.33 ± 18.47

202.31 ± 19.43

90.21 ± 12.57

134.37 ± 11.18 172.12 ± 10.29

352

28.95

5.90

–6.69

–0.69

–16.55

–3.16

–12.62

–7.83

Heel Strike

Toe-off

Stride Time

Stance Time

Swing Time

Step Time

Stride Length

Stride Velocity

True error

8.22

9.79

2.76

7.91

4.60

1.64

0.57

0.51

Error (%)

Normal

0.93

0.92

0.97

0.89

0.84

0.98

—

—

ICC(2,k)

17.90

10.58

–3.94

–24.99

–0.91

–19.01

0.63

31.36

True error

8.77

6.34

3.24

8.76

4.92

2.98

0.60

0.62

Error (%)

Fast

0.97

0.97

0.94

0.84

0.87

0.91

—

—

ICC(2,k)

–6.40

–11.93

–2.84

–7.88

7.52

–11.95

4.87

41.38

True error

10.33

11.53

3.54

7.51

5.37

1.93

0.56

0.60

Error (%)

Slow

0.91

0.88

0.93

0.87

0.85

0.96

—

—

ICC(2,k)

Table 2  Mean values, taken across all subjects for true error, percentage error and ICC(2,k) values for each speed and each gait parameter. True error is measured in ms for temporal gait parameters, cm for stride length and cm/s for stride velocity.

Body-Worn Sensors and the GAITRite Walkway   353

Figure 2 — True error (in ms) between body-worn sensor algorithms and GAITRite for HS and TO points at three different self-selected walking speeds. Negative values of true error indicate that a HS or TO point was detected by the body-worn sensor algorithm before detection by GAITRite. Median values are indicated by the central line, while the rectangular box represents the interquartile range (25th to 75th percentiles). Outliers are indicated by “+”, while the whiskers extend to the most extreme data points not considered outliers.

Discussion This study demonstrates good concurrent validity of shank mounted body-worn gyroscopes when measuring spatiotemporal gait parameters during walking at three different speeds in healthy subjects, compared with the GAITRite electronic walkway. GAITRite is a widely used clinical gait analysis tool that is not confined to custombuilt gait analysis laboratories. Establishing concurrent validity between GAITRite and the gait analysis platform and algorithms presented here is an important step in determining the utility of this method in an environment that is more accessible to clinicians. Previous studies have used GAITRite for the measurement of both averaged and individual step parameters of gait (Hartmann et al., 2009; Webster et al., 2005). When average step data are used for comparison between two systems, greater levels of agreement have been shown. For example, Hartmann et al. (Hartmann et al., 2009), compared the DynaPort accelerometer-based system mounted on the trunk with GAITRite for calculation of a number of spatiotemporal parameters and found moderate levels of agreement for individual step duration—ICCs: 0.81–0.88 for the three speeds—compared with ICCs of 1.0 for averaged step data. In this study, individual footsteps for each walking speed were used to determine the concurrent validity. This is an important characteristic of this study as ICCs for individual step and stride times ranged from 0.91 to 0.98 across all speeds, indicating that this system can facilitate accurate assessment of temporal

stride-to-stride fluctuations. This is a valuable capability of the system as it can provide information about the temporal structure or organization of locomotion over time, which is currently an important subject in the field of gait analysis. Other studies have used GAITRite to validate onbody sensor technologies in certain gait parameters. A study by Najafi et al. (2009) used body-worn sensors containing gyroscopes and a Physilog ambulatory datalogger (BioAGM, Switzerland). The gyroscopes were used to calculate spatiotemporal parameters of gait, and compared with the GAITRite mat. Agreement between GAITRite and the Physilog system was high with respect to the gait cycle time (GCT, equivalent to stride time in this study), with an average difference of –12 ms and standard deviation of less than 46 ms, across all three speeds. The results for stride time in the current study demonstrated a similar accuracy and improved precision, relative to the Physiolog data measurement, with an average difference of –12.55 ± 11.48 ms across three speeds. The aim of method comparison studies is to establish the agreement between methods or instruments that are measuring the same task. The results then need to be considered within a clinical context to determine their implications. With this in mind, the magnitude of error for stride time found in this study was compared with a clinically meaningful magnitude of error reported in a previous study carried out by Stokic et al. (2009), where spatiotemporal parameters from healthy and poststroke subjects were compared using the GAITRite and a motion

354  Greene et al.

capture system. The authors reported differences in stride time between patient groups to be 550 ms. The maximum difference in stride time, as measured with inertial sensors in this study, compared with the GAITRite was approximately 20 ms. We can therefore infer that the differences observed in measurement systems would not affect the clinical interpretation of such results. Gait analysis is frequently used in clinical decision making. Therefore, it is essential that the validity of any gait analysis system be firmly established before it is used in clinical situations. The gyroscope-based algorithms and gait analysis system reported here compare favorably with the GAITRite electronic walkway in terms of the error and intraclass correlation coefficients at three different self-selected walking speeds for normal healthy subjects. The ICC for all temporal gait parameters was found to be greater than 0.84, indicating good agreement. A limitation of the current study is that the sampling rates from the gyroscopes and GAITRite were not matched, which may have given rise to errors in the comparison. In addition, a limitation of the algorithms for body-worn sensors employed in the current study is that parameters examining base of support, such as step width and step length, cannot be readily obtained. One disadvantage of GAITRite is that data capture is restricted to a few steps at a time, which may not be representative of the person’s real gait. True measurement of gait ideally requires large amounts of data in steady-state walking, in a natural environment. Owings and Grabiner (Owings et al., 2003) suggested that accurate estimation of gait variability requires at least 400 steps. The gait analysis system presented can potentially address these requirements, thereby creating a more flexible and instructive method for examination of locomotion. Acknowledgments This research was completed as part of a wider program of research within the TRIL Centre (Technology Research for Independent Living; http://www.trilcentre.org.). The TRIL Centre is a multidisciplinary research centre, bringing together researchers from UCD, TCD, NUIG, Intel, and GE Healthcare, funded by Intel, GE Healthcare and IDA Ireland.

References Aminian, K., Najafi, B., Büla, C., Leyvraz, P.F., & Robert, P. (2002). Spatio-temporal parameters of gait measured by an ambulatory system using miniature gyroscopes. Journal of Biomechanics, 35, 689–699. PubMed doi:10.1016/ S0021-9290(02)00008-8 Bilney, B., Morris, M., & Webster, K. (2003). Concurrent related validity of the GAITRite walkway system for quantification of the spatial and temporal parameters of gait. Gait & Posture, 17, 68–74. PubMed doi:10.1016/ S0966-6362(02)00053-X Burns, A., Greene, B.R., McGrath, M.J., O’Shea, T.J., Kuris, B., Ayer, S.M., . . .. (2010). SHIMMER - A Wireless Sensor Platform for Non-invasive Biomedical Research. IEEE Sensors, 10, 1527–1534. doi:10.1109/JSEN.2010.2045498

Delahunt, E., Monaghan, K., & Caulfield, B. (2006). Altered neuromuscular control and ankle joint kinematics during walking in subjects with functional instability of the ankle joint. American Journal of Sports Medicine, 34, 1970–1976. PubMed doi:10.1177/0363546506290989 Desailly, E., Daniel, Y., Sardain, P., & Lacouture, P. (2009). Foot contact event detection using kinematic data in cerebral palsy children and normal adults gait. Gait & Posture, 29, 76–80. PubMed doi:10.1016/j.gaitpost.2008.06.009 Doheny, E.P., Foran, T., & Greene, B.R. (2010). A single gyroscope method for spatial gait analysis. Paper presented at the 32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Society. Ferraris, F., Grimaldi, U., & Parvis, M. (1995). Procedure for effortless in-field calibration of three-axis rate gyros and accelerometers. Sensors and Materials, 7, 311–330. Greene, B.R., McGrath, D., O’Neill, R., O’Donovan, K.J., Burns, A., & Caulfield, B. (2010a). Adaptive estimation of temporal gait parameters using body-worn gyroscopes, 32nd IEEE Engineering in Medicine and Biology conference. Buenos Aires, Argentina. Greene, B.R., McGrath, D., O’Neill, R., O’Donovan, K.J., Burns, A., & Caulfield, B. (2010b). An adaptive gyroscope based algorithm for temporal gait analysis. Medical & Biological Engineering & Computing, 48, 1251–1260. PubMed doi:10.1007/s11517-010-0692-0 Hartmann, A., Luzi, S., Murer, K., de Bie, R.A., & de Bruin, E.D. (2009). Concurrent validity of a trunk tri-axial accelerometer system for gait analysis in older adults. Gait & Posture, 29, 444–448. PubMed doi:10.1016/j. gaitpost.2008.11.003 Jasiewicz, J.M., Allum, J.H.J., Middleton, J.W., Barriskill, A., Condie, P., Purcell, B., . . .. (2006). Gait event detection using linear accelerometers or angular velocity transducers in able-bodied and spinal-cord injured individuals. Gait & Posture, 24, 502–509. PubMed doi:10.1016/j. gaitpost.2005.12.017 Kressig, R.W., & Beauchet, O. (2006). Guidelines for clinical applications of spatio-temporal gait analysis in older adults. Aging Clinical and Experimental Research, 18, 174–176. PubMed McGrath, M. J., & Dishongh, T. J. (2009). A Common Personal Health Research Platform—SHIMMER and BioMOBIUS. Intel technology journal, 13, 122–147. Najafi, B., Aminian, K., Loew, F., Blanc, Y., & Robert, P.A. (2002). Measurement of stand-sit and sit-stand transitions using a miniature gyroscope and its application in fall risk evaluation in the elderly. IEEE Transactions on BioMedical Engineering, 49, 843–851. PubMed doi:10.1109/ TBME.2002.800763 Najafi, B., Helbostad, J.L., Moe-Nilssen, R., Zijlstra, W., & Aminian, K. (2009). Does walking strategy in older people change as a function of walking distance. Gait & Posture, 29, 261–266. PubMed doi:10.1016/j.gaitpost.2008.09.002 Owings, T.M., & Grabiner, M.D. (2003). Measuring step kinematic variability on an instrumented treadmill: how many steps are enough? Journal of Biomechanics, 36, 662–666. PubMed doi:10.1016/S0021-9290(03)00108-8 Paul, S.M., Siegel, K.L., Malley, J., & Jaeger, R.J. (2007). Evaluating interventions to improve gait in cerebral palsy: a meta-analysis of spatiotemporal measures. Developmental Medicine and Child Neurology, 49, 542–549. PubMed doi:10.1111/j.1469-8749.2007.00542.x Sabatini, A.M., Martelloni, C., Scapellato, S., & Cavallo, F. (2005). Assessment of walking features from foot inertial

Body-Worn Sensors and the GAITRite Walkway   355

sensing. IEEE Transactions on Biomedical Engineering, 52, 486–494. PubMed doi:10.1109/TBME.2004.840727 Shrader, M.W., Bhowmik-Stoker, M., Jacofsky, M.C., & Jacofsky, D.J. (2009). Gait and stair function in total and resurfacing hip arthroplasty: a pilot study. Clinical Orthopaedics and Related Research, 467, 1476–1484. PubMed doi:10.1007/s11999-009-0791-0 Shrout, P.E., & Fleiss, J.L. (1979). Intraclass correlations: Uses in assessing rater reliability. Psychological Bulletin, 86, 420–428. PubMed doi:10.1037/0033-2909.86.2.420 Stokic, D. S., Horn, T. S., Ramshur, J. M., & J.W., C. (2009). Agreement between temporospatial gait parameters of an electric walkway and a motion capture system in healthy and chronic stroke populations. American Journal of Physical Medicine & Rehabilitation, 88, 437–444. PubMed doi:10.1097/PHM.0b013e3181a5b1ec

Tong, K., & Granat, M.H. (1999). A practical gait analysis system using gyroscopes. Medical Engineering & Physics, 21, 87–94. PubMed doi:10.1016/S1350-4533(99)00030-2 Webster, K.E., Wittwer, J.E., & Feller, J.A. (2005). Validity of the GAITRite walkway system for the measurement of averaged and individual step parameters of gait. Gait & Posture, 22, 317–321. PubMed doi:10.1016/j.gaitpost.2004.10.005 Youdas, J.W., Childs, K.B., McNeil, M.L., Mueller, A.C., Quilter, C.M., & Hollman, J.H. (2010). Responsiveness of 2 procedures for measurement of temporal and spatial gait parameters in older adults. PM&R, 2, 537–543. Zijlstra, W., & Hof, A.L. (2003). Assessment of spatio-temporal gait parameters from trunk accelerations during human walking. Gait & Posture, 18, 1–10. PubMed doi:10.1016/ S0966-6362(02)00190-X