- Email: [email protected]
Validation of a commercial inertial sensor system for spatiotemporal gait measurements in children
Gait & Posture 51 (2017) 14–19
Contents lists available at ScienceDirect
Gait & Posture journal homepage: www.elsevier.com/locate/gaitpost
Full length article
Validation of a commercial inertial sensor system for spatiotemporal gait measurements in children Joel L. Lanovaza,* , Alison R. Oatesa , Tanner T. Treena , Janelle Ungera , Kristin E. Musselmanb,c,d a
College of Kinesiology, University of Saskatchewan, 87 Campus Drive, Saskatoon, Saskatchewan S7N 5B2, Canada SCI Mobility Lab., Lyndhurst Centre, Toronto Rehabilitation Institute—University Health Network, 520 Sutherland Drive, Toronto, Ontario M4G 3V9, Canada Department of Physical Therapy, Faculty of Medicine, University of Toronto, Toronto, Ontario, Canada d School of Physical Therapy, College of Medicine, University of Saskatchewan, Saskatoon, Saskatchewan, Canada b c
A R T I C L E I N F O
Article history: Received 16 April 2016 Received in revised form 15 September 2016 Accepted 24 September 2016 Keywords: Gait Inertial sensor 3D motion capture Children Spatiotemporal Validation
A B S T R A C T
Although inertial sensor systems are becoming a popular tool for gait analysis in both healthy and pathological adult populations, there are currently no data on the validity of these systems for use with children. The purpose of this study was to validate spatiotemporal data from a commercial inertial sensor system (MobilityLab) in typically-developing children. Data from 10 children (5 males; 3.0–8.3 years, mean = 5.1) were collected simultaneously from MobilityLab and 3D motion capture during gait at selfselected and fast walking speeds. Spatiotemporal parameters were compared between the two methods using a Bland-Altman method. The results indicate that, while the temporal gait measurements were similar between the two systems, MobilityLab demonstrated a consistent bias with respect to measurement of the spatial data (stride length). This error is likely due to differences in relative leg length and gait characteristics in children compared to the MobilityLab adult reference population used to develop the stride length algorithm. A regression-based equation was developed based on the current data to correct the MobilityLab stride length output. The correction was based on leg length, stride time, and shank range-of-motion, each of which were independently associated with stride length. Once the correction was applied, all of the spatiotemporal parameters evaluated showed good agreement. The results of this study indicate that MobilityLab is a valid tool for gait analysis in typically-developing children. Further research is needed to determine the efficacy of this system for use in children suffering from pathologies that impact gait mechanics. ã 2016 Elsevier B.V. All rights reserved.
1. Introduction The past several years have seen increased use of inertial sensors to analyze movement in laboratory, clinic, and daily living environments [1]. By utilizing accelerometers, gyroscopes, or magnetometers (or a combination of these), inertial sensors can provide a wealth of data regarding the characteristics of global and segment-specific movement during a variety of tasks. Additionally, the sensors and recording equipment are relatively compact, portable, and low cost compared to traditional laboratory-based equipment (such as multi-camera 3D motion capture or instrumented mats), and can be used to collect human movement data in
* Corresponding author. E-mail address: [email protected] (J.L. Lanovaz). http://dx.doi.org/10.1016/j.gaitpost.2016.09.021 0966-6362/ã 2016 Elsevier B.V. All rights reserved.
environments and contexts where the use of traditional equipment is not possible. Inertial sensor technology that can be used in both laboratory and clinical environments has the potential to be a widely applicable method for researchers and clinicians to evaluate gait in a variety of healthy and clinical populations. One widely-used inertial sensor system is the MobilityLab system (APDM, Portland, OR). This system utilizes six inertial sensors, each containing tri-axial accelerometers, gyroscopes, and magnetometers providing a comprehensive evaluation of the spatiotemporal characteristics of motion during a variety of preprogrammed testing protocols [2–4]. Data collected from these sensors is transmitted wirelessly to a software program, which uses algorithms based on aggregated reference data that have been validated against both 3D motion capture and force plate data to calculate the spatiotemporal characteristics (such as stride time, stride length, and velocity of each stride) of movement [3–5]. The
J.L. Lanovaz et al. / Gait & Posture 51 (2017) 14–19
system is also capable of discriminating between different movements associated with various mobility tests such as the sit-to-stand and turning phases of the timed-up-and-go test [6], and has been used in the evaluation of gait and mobility in clinical populations including persons with Parkinson’s disease and multiple sclerosis [7,8]. While previous research has indicated that inertial sensor systems such as MobilityLab are a valid and reliable method of analyzing movement in adults [9], there has yet to be any research on the validity of their use in children. Since achieving functional gait and maximizing ambulatory independence are two of the most important functional outcomes for children suffering from musculoskeletal and neurological pathologies [10], it is crucial for clinicians to be able to analyze gait in children to recognize and attempt to correct any impairments and sub-optimal movement patterns that may be limiting functional capacity. Compared to traditional measurement tools used for gait analysis, inertial sensors offer several distinct benefits when working with children. The sensors are much easier to don and doff than reflective marker sets and use Velcro straps rather than adhesives, reducing the chances of skin irritation and/or discomfort during removal. Additionally, while most methods of gait analysis restrict movement to a given space or require the child to contact a target with their foot, the sensors allow the child to walk using their normal movement pattern with no environmental constraints. While some research exists evaluating the use of inertial sensors as a tool for gait analysis in children with cerebral palsy [11–13], there are currently no data evaluating the validity of inertial sensor systems relative to 3D motion capture (the gold standard of gait analysis). Direct measurements of kinematic parameters like linear acceleration and angular velocity from inertial systems are fairly accurate; however indirect measures such as spatiotemporal parameters often rely on algorithms with assumptions and reference values based on adult data. It is unclear if these approaches will result in accurate data when applied to children. The objective of this study is to validate the use of the MobilityLab inertial sensor system to obtain spatiotemporal parameters of gait in typically-developing children by comparing the level of agreement between data from the sensors and those obtained via 3D motion capture. We hypothesize that temporal data based on event detection will be accurate but estimations of spatial data may be influenced by adult-data assumptions inherent to the MobilityLab algorithms. 2. Methods Ten typically-developing children (five males) participated in the study (mean age 5.1 yrs, range 3.0 yrs–8.3 yrs). Participants were eligible for the study if they were between the ages of three and 10, free of any neurological disorders or lower limb musculoskeletal injuries, and were full term (37 weeks gestational age) at birth. The study was approved by the institutional Research Ethics Board and informed consent was obtained from the children’s guardians. In addition to obtaining informed consent from each child’s guardian, all of the participants gave verbal assent prior to their involvement in the study. Data were collected from each child as they walked in a straight line along a 7 m long walkway. Each child performed six to eight walking trials with approximately half the trials at a self-selected velocity. In the other half of the trials, the child was instructed to walk faster without running. The result was a range of walking velocities, with an overall mean velocity of 1.07 m/s, an average minimum of 0.83 m/s (SD 0.18 m/s) and an average maximum 1.51 m/s (SD 0.24 m/s). Encouragement was provided as needed to
15
maintain the child’s attention and engagement but the child walked without any hands-on assistance. Height was obtained to the nearest 0.1 cm using a stadiometer (mean 106.8 cm, range 93.5 cm–118.0 cm). Leg length (mean 42.7%, range 37.5 to 47.8%) was estimated as a percentage of total height from the motion capture data and was defined as the vertical length from the greater trochanter to the ankle (averaged across both legs) during standing. The MobilityLab system (version 1.0.0.201503302135) was used to collect the inertial-based spatiotemporal and kinematic data. Each child wore a total of six inertial sensors positioned on the dorsal side of both wrists, on the sternum close to the clavicular notch, on the lower back in correspondence to L4/L5, and on the frontal side of the shanks close to the malleoli. Data were collected wirelessly at a sampling rate of 128 Hz using MobilityLab’s iWalk module, which is designed for straight line walking of indeterminate length. Simultaneously, reference kinematic data were collected at 100 Hz using an 8 camera 3D motion capture system (Vicon Nexus, Centennial, CO). Reflective tracking markers (14 mm diameter) were fixed to the areas of the greater trochanter, lateral femoral condyle, lateral malleoli, heel and toe of both legs. Additionally, three tracking markers (9 mm diameter) were fixed to the inertial sensor on the sternum with two markers in line with a sensor axis and the third defining a cardinal plane in sensor coordinate system. Foot contact and lift off were detected from the motion capture data using a manually-tuned automated foot velocity threshold algorithm [14,15] and were used to calculate temporal data. Spatial data (i.e. stride length) were calculated using the heel position data. Inertial and motion capture data were synchronized during post-processing using custom software (Matlab R2006b, Mathworks, Natick, MA). Three dimensional acceleration of the sternum inertial sensor was calculated using motion capture data and transformed into the sensor coordinate system. These data were then time-matched to within 0.01 s of the corresponding raw sensor accelerometer data using a custom semi-automatic correlation method which used cross-correlation to provide an initial guess and then manual adjustment to find the final synchronization point. Data from six strides from each of the right and left legs were randomly chosen from each participant resulting in a total of 120 strides used for analysis. Four main spatiotemporal variables were compared between the systems; stride time (StrT), stance time (StnT), stride length (StrL) and stride velocity (StrV). StrT was defined as the time from heel strike on one foot to the next heel strike of the same foot. StrV was calculated on a stride-by-stride basis as the ratio of StrL and StrT. StnT was assessed in addition to StrT as both heel strike and toe off identification are included in StnT calculations. According to the MobilityLab manufacturer [16], the procedures used to generate temporal and stride length data from the inertial sensors are based on published algorithms [3,5]. For this study, the video-based motion capture system was assumed to be the gold standard. Data were compared between the two methods using the Bland-Altman method [17,18]. The BlandAltman method allows for comparisons between two different measurement systems to assess agreement when measuring the same set of data. The method provides an estimate of the bias between the systems and a measure of agreement known as limits of agreement (LoA). Since the magnitude of the some of the variables appeared to influence the bias, bias values and LoA were tested for significant non-zero slopes [18] and, when this was found, were fitted using linear regression [18]. Additionally, mean differences and root mean square (RMS) differences between the two systems were calculated. All statistics were calculated using SPSS (Version 23, IBM Corp, Armonk, NY).
16
J.L. Lanovaz et al. / Gait & Posture 51 (2017) 14–19
Fig. 1. Bland-Altman plots comparing MobilityLab (ML) and 3D motion capture (Vicon) results for (A) Stance time, (B) Stride time, (C) Stride length and (D) Stride velocity. Bias (solid line) and limits of agreement (dashed lines) are shown for each variable.
3. Results StnT and StrT were comparable between systems with little bias (Fig. 1). The LoA for StnT and StrT were approximately +/ 0.05 s and +/ 0.02 s respectively (Fig. 1). Even though the bias and LoA for the StrT were not constant, they were approximately equal to two motion capture sample point intervals. Examination of the raw foot fall data showed that 93% of initial contact and 68% of toe off identifications from the inertial system were within one frame of the motion capture system. StrL showed a large, non-constant bias ranging from +10%h to 10%h which indicated that MobilityLab overestimated the length of shorter strides and underestimated longer strides (Fig. 1). These differences would lead to 9–14% errors in StrL estimations at the
slowest and fastest walking velocities. StrV also showed a nonconstant bias with similar over and under estimation (Fig. 1) and a widening LoA indicating that agreement was worse at higher velocities. The errors in StrV were likely largely derived from StrL errors as the StrT differences were quite small. Mean and RMS differences are provided in Table 1. The error in StrL estimation for children using the MobilityLab algorithm was not unexpected. The StrL algorithm requires LL as well as the sagittal plane orientation of the shank and thigh at heel strike and toe off [5]. MobilityLab uses a value for LL that is a fixed percentage of total height [16] based on published data for adults (49.1%, [19] as referenced in [20]). LL in children is generally a smaller proportion of height as compared to adults [21] which was true in the current study (mean 42.7%). Additionally, the standard
Table 1 Mean and RMS differences between the two measurement systems. A positive value for the mean difference indicates that the MobilityLab values overestimate the motion capture data. Differences after the applied correction for StrL and StrV are also given. Original Data
StrT (s) StnT (s) StrL (%h) StrV (%h/s)
After Correction
Mean Difference
RMS Difference
0.001 s 0.003 s 4.68%h 4.97%h/s
0.014 s 0.026 s 10.30%h 12.10%h/s
%h indicates percent of the participant’s height.
Mean Difference
0.04%h 0.09%h
RMS Difference
6.43%h 7.80%h
J.L. Lanovaz et al. / Gait & Posture 51 (2017) 14–19
17
Table 2 Results for the step-wise linear regression modeling of the StrL error. The data from each step is given along with the subsequent R2 change. Model
Overall Adjusted R2
R2 change
Independent variables
Partial R2
b
Model 1 Model 2
0.22 0.35
n/a 0.14
Model 3
0.53
0.18
ShkROM ShkROM LL ShkROM LL StrT
0.22 0.23 0.18 0.30 0.30 0.29
0.48a 0.44a 0.37a 0.44a 0.45a 0.43a
a
Indicates significance at p < 0.001.
MobilityLab sensor configuration only collects angular motion data from the shank and uses a prediction formula, also based on adult data, to estimate the thigh position [3]; therefore, it was expected that the StrL error would be related to both LL and shank movement and would also likely be a function of walking speed, which would be reflected in StrT. We assessed the nature of the StrL error with bivariate correlations which showed that, on a stride-by-stride basis, StrL error was significantly correlated with shank range of motion (ShkROM) (r = 0.478, p < 0.001) and StrT (r = 0.334, p < 0.001) from MobilityLab. LL was also significantly correlated to average StrL error (r = 0.422, p < 0.001). ShkROM and StrT were considered surrogates for shank position and walking speed respectively. These values are reliable and are readily available from the standard MobilityLab output. ShkROM and StrT were not significantly correlated to each other while LL was minimally correlated with StrT (r = 0.184, p = 0.047) and was not correlated with ShkROM. A step-wise linear regression was used to model StrL error. Independent variables included ShkROM, StrT and LL. Results showed all three variables were significant in the model (Table 2). The corresponding regression formula (Eq. (1)) was then used to estimate StrL error for each stride: StrL error = 82.399
0.420 ShkROM
1.460 LL + 24.715 StrT(1)
where ShkROM is expressed in degrees, LL is in percent of total height, StrT is in seconds and StrL error is in percent of total height. The error correction was added to the original MobilityLab StrL to generate a corrected StrL value. The MobilityLab StrV was then recalculated using the corrected StrL. The corresponding BlandAltman plots showed greatly improved biases and a generally smaller LoA for StrV, although the StrL agreement tended to get worse at longer stride lengths (Fig. 2).
The mean StrL error was reduced from 4.7%h to 0.04%h and the RMS StrL error dropped from 10.3%h to 6.4%h after applying the correction. Similarly, the mean StrV error was reduced from 5.0%h/s to 0.1%h/s and the RMS StrV error dropped from 12.1%h/s to 7.8%h/s with the correction (Table 1). 4. Discussion This study examined the accuracy of MobilityLab, a commercial inertial sensor system, for analyzing basic spatiotemporal gait parameters in young children. MobilityLab produced accurate results for temporal gait variables which were comparable to what can be obtained with a 3D motion capture system. For the spatial variable of StrL, the MobilityLab algorithm overestimated short strides and underestimated long strides. A regression analysis was used to generate a StrL correction formula based on measured factors known to influence the MobilityLab algorithm. Improvements to MobilityLab StrL and StrV accuracy were seen with the application of the correction. The MobilityLab system was capable of detecting foot falls with a high degree of accuracy when compared to 3D motion capture. The LoA for StrT were approximately 2 motion capture frames intervals which is well within the normal error range of almost all foot fall detection algorithms [14,22–29]. The LoA for StnT were higher than for StrT which was likely a result of higher variability in detecting toe off events which is typical of automatic detection approaches [15]. The detection algorithm chosen for the 3D motion capture data in this study is just one of many available algorithms [15] and it is possible that a different approach would produce different values. Estimating StrL from a limited set of inertial sensors is challenging. The StrL algorithm employed in MobilityLab [5] was developed and tested for adults and, as such, has certain built
Fig. 2. Bland-Altman plots comparing corrected MobilityLab (ML) with 3D motion capture data (Vicon) for (A) Stride length and (B) Stride velocity. ML data were corrected using Eq. (1). Bias (solid lines) and limits of agreement (dashed lines) are shown.
18
J.L. Lanovaz et al. / Gait & Posture 51 (2017) 14–19
in assumptions which affect its performance when applied to children. The first assumption is that of fixed leg length proportionally with respect to height based on adult data. All of the children in this study had a LL that was a smaller proportion of their height than what is seen in a typical adult. Based on sample calculations we performed using the published algorithm [5], when LL is overestimated (i.e. using adult proportions rather than child), StrL will be consistently overestimated. This was the case for 70% of the StrL data in this study; however, since not all StrL were overestimated, other factors, such as limb kinematics, are also likely involved. MobilityLab is able to use a reduced set of sensors to analyze gait by utilizing an algorithm which estimates thigh angle based on shank data [3]. The underlying data structure for this estimation process was derived using data from adults walking on a treadmill [3]. Movement patterns in children can be significantly different than adults [30] and this likely affects the performance of the MobilityLab algorithm. The fact that ShkROM was one of the most significant predictors of StrL error in this study supports this claim. Stride velocity also appears to effect StrL estimation from MobilityLab as evidenced by the significant contribution of StrT as a StrL error predictor. There is likely a complex interaction between shank and thigh movement and walking velocity which would require specific calibration with children’s gait, similar to what was done in adults for the original algorithm [3]. The use of the correction equation presented in this paper significantly improved the StrL and StrV estimations from MobilityLab; mean StrL and StrV errors became negligible and absolute differences were reduced by an average of 40%. This indicates that, after the correction is applied, MobilityLab may be considered a valid substitute for a 3D motion-capture system at predicting the spatiotemporal parameters of gait in typicallydeveloping children. Additionally, since the factors included in the correction equation (ShkROM, StrT, and LL) are automatically generated from the MobilityLab data processing software, they are easy to access and enter into the correction equation after analysis. The limitations to this study included the low sample size, which makes it difficult to generalize the results of the regression. Since there were only 5 participants for each sex ranging from 3 to 10 years old, the descriptive data was highly variable. Additionally, body proportions and gait maturity develop and change throughout childhood and adolescence, which likely contributed to the substantial variation in descriptive parameters between participants [21]. The variables used in the regression were purposefully selected based on the known factors related to the calculation of StrL in the MobilityLab algorithm. The low sample size precluded a wide exploration of other possible factors but it is likely that the chosen variables are among the most influential. Although the potential for use of the MobilityLab system for use in children with pathological conditions affecting gait exists, further research is needed to determine if this system can accurately identify changes in spatiotemporal parameters in children with conditions/impairments that impact walking. 5. Conclusion Although previous studies have confirmed the validity of inertial sensors for use during gait analysis in adults, our study represents the first instance where the validity of inertial sensor measurements have been evaluated in children. Although we found the MobilityLab system to be a valid measure for temporal gait parameters and event detection, the system demonstrated a consistent bias with respect to StrL. A regression equation was used to correct for the observed bias, resulting in a substantial improvement in agreement between the MobilityLab and 3D motion capture systems with respect to StrL and StrV. The results
indicate that MobilityLab can be a valid tool for gait analysis in typically-developing children but that care must be taken when applying stride length algorithms developed for adult populations. Acknowledgements Funding for this study was provided by the Spina Bifida and Hydrocephalus Association of Canada. The sponsor was not directly involved in the design or the study, the collection/ analysis/interpretation of data, or the production of this manuscript. The authors would like to thank all of the participants. Conflict of interest The authors have no conflicts of interest. References [1] P.B. Shull, W. Jirattigalachote, M.A. Hunt, M.R. Cutkosky, S.L. Delp, Quantified self and human movement: a review on the clinical impact of wearable sensing and feedback for gait analysis and intervention, Gait Posture 40 (2014) 11–19. [2] A. Salarian, F.B. Horak, C. Zampieri, P. Carlson-Kuhta, J.G. Nutt, K. Aminian, iTUG, a sensitive and reliable measure of mobility, IEEE Trans. Neural Syst. Rehabil. Eng. 18 (2010) 303–310. [3] A. Salarian, P.R. Burkhard, F.J. Vingerhoets, B.M. Jolles, K. Aminian, A novel approach to reducing number of sensing units for wearable gait analysis systems, IEEE Trans. Biomed. Eng. 60 (2013) 72–77. [4] M. Mancini, L. King, A. Salarian, L. Holmstrom, J. McNames, F.B. Horak, Mobility lab to assess balance and gait with synchronized body-worn sensors, J. Bioeng. Biomed. Sci. (Suppl. (1)) (2011) 007. [5] K. Aminian, B. Najafi, C. Büla, P. Leyvraz, P. Robert, Spatio-temporal parameters of gait measured by an ambulatory system using miniature gyroscopes, J. Biomech. 35 (2002) 689–699. [6] J.T. Coulthard, T.T. Treen, A.R. Oates, J.L. Lanovaz, Evaluation of an inertial sensor system for analysis of timed-up-and-go under dual-task demands, Gait Posture 41 (2015) 882–887. [7] C. Zampieri, A. Salarian, P. Carlson-Kuhta, K. Aminian, J.G. Nutt, F.B. Horak, The instrumented timed up and go test: potential outcome measure for disease modifying therapies in Parkinson's disease, J. Neurol. Neurosurg. Psychiatry 81 (2010) 171–176. [8] R.I. Spain, R.J. St George, A. Salarian, M. Mancini, J.M. Wagner, F.B. Horak, D. Bourdette, Body-worn motion sensors detect balance and gait deficits in people with multiple sclerosis who have normal walking speed, Gait Posture 35 (2012) 573–578. [9] A. Muro-de-la-Herran, B. Garcia-Zapirain, A. Mendez-Zorrilla, Gait analysis methods: an overview of wearable and non-wearable systems, highlighting clinical applications, Sensors 14 (2014) 3362–3394. [10] S.K. Campbell, R.J. Palisano, D.W. Vander Linden, Physical Therapy for Children, Elsevier Saunders, 2006. [11] A.G. Cutti, A. Ferrari, P. Garofalo, M. Raggi, A. Cappello, A. Ferrari, ‘Outwalk’: a protocol for clinical gait analysis based on inertial and magnetic sensors, Med. Biol. Eng. Comput. 48 (2010) 17–25. [12] C. van den Noort Josien, A. Ferrari, A.G. Cutti, J.G. Becher, J. Harlaar, Gait analysis in children with cerebral palsy via inertial and magnetic sensors, Med. Biol. Eng. Comput. 51 (2013) 377–386. [13] A.B. Bourgeois, B. Mariani, K. Aminian, P. Zambelli, C. Newman, Spatiotemporal gait analysis in children with cerebral palsy using, foot-worn inertial sensors, Gait Posture 39 (2014) 436–442. [14] S. Ghoussayni, C. Stevens, S. Durham, D. Ewins, Assessment and validation of a simple automated method for the detection of gait events and intervals, Gait Posture 20 (2004) 266–272. [15] D.A. Bruening, S.T. Ridge, Automated event detection algorithms in pathological gait, Gait Posture 39 (2014) 472–477. [16] M. Johnson, APDM Inc., Personal communication. [17] J.M. Bland, D. Altman, Statistical methods for assessing agreement between two methods of clinical measurement, Lancet 327 (1986) 307–310. [18] J.M. Bland, D.G. Altman, Measuring agreement in method comparison studies, Stat. Methods Med. Res. 8 (1999) 135–160. [19] R. Drillis, R. Contini, Body Segment Parameters Report No 1163-03 Office of Vocational Rehabilitation, Department of health, Education and Welfare, New York, 1966. [20] D.A. Winter, Biomechanics and Motor Control of Human Movement, John Wiley & Sons, 2009. [21] B. Bogin, M.I. Varela-Silva, Leg length, body proportion, and health: a review with a note on beauty, Int. J. Environ. Res. Public Health 7 (2010) 1047–1075. [22] A. De Asha, M. Robinson, G. Barton, A marker based kinematic method of identifying initial contact during gait suitable for use in real-time visual feedback applications, Gait Posture 36 (2012) 650–652.
J.L. Lanovaz et al. / Gait & Posture 51 (2017) 14–19 [23] E. Desailly, Y. Daniel, P. Sardain, P. Lacouture, Foot contact event detection using kinematic data in cerebral palsy children and normal adults gait, Gait Posture 29 (2009) 76–80. [24] A. Hreljac, R.N. Marshall, Algorithms to determine event timing during normal walking using kinematic data, J. Biomech. 33 (2000) 783–786. [25] B. Hsue, F. Miller, F. Su, J. Henley, C. Church, Gait timing event determination using kinematic data for the toe walking children with cerebreal palsy, J. Biomech. 40 (2007) S529. [26] J.M. Jasiewicz, J.H. Allum, J.W. Middleton, A. Barriskill, P. Condie, B. Purcell, R.C. T. Li, Gait event detection using linear accelerometers or angular velocity transducers in able-bodied and spinal-cord injured individuals, Gait Posture 24 (2006) 502–509.
19
[27] C.M. O’Connor, S.K. Thorpe, M.J. O’Malley, C.L. Vaughan, Automatic detection of gait events using kinematic data, Gait Posture 25 (2007) 469–474. [28] J. Salazar-Torres, Validity of an automated gait event detection algorithm in children with cerebral palsy and non-impaired children, Gait Posture 24 (2006) S130–1. [29] J. Zeni, J. Richards, J. Higginson, Two simple methods for determining gait events during treadmill and overground walking using kinematic data, Gait Posture 27 (2008) 710–714. [30] A.W. Froehle, R.W. Nahhas, R.J. Sherwood, D.L. Duren, Age-related changes in spatiotemporal characteristics of gait accompany ongoing lower limb linear growth in late childhood and early adolescence, Gait Posture 38 (2013) 14–19.
Contents lists available at ScienceDirect
Gait & Posture journal homepage: www.elsevier.com/locate/gaitpost
Full length article
Validation of a commercial inertial sensor system for spatiotemporal gait measurements in children Joel L. Lanovaza,* , Alison R. Oatesa , Tanner T. Treena , Janelle Ungera , Kristin E. Musselmanb,c,d a
College of Kinesiology, University of Saskatchewan, 87 Campus Drive, Saskatoon, Saskatchewan S7N 5B2, Canada SCI Mobility Lab., Lyndhurst Centre, Toronto Rehabilitation Institute—University Health Network, 520 Sutherland Drive, Toronto, Ontario M4G 3V9, Canada Department of Physical Therapy, Faculty of Medicine, University of Toronto, Toronto, Ontario, Canada d School of Physical Therapy, College of Medicine, University of Saskatchewan, Saskatoon, Saskatchewan, Canada b c
A R T I C L E I N F O
Article history: Received 16 April 2016 Received in revised form 15 September 2016 Accepted 24 September 2016 Keywords: Gait Inertial sensor 3D motion capture Children Spatiotemporal Validation
A B S T R A C T
Although inertial sensor systems are becoming a popular tool for gait analysis in both healthy and pathological adult populations, there are currently no data on the validity of these systems for use with children. The purpose of this study was to validate spatiotemporal data from a commercial inertial sensor system (MobilityLab) in typically-developing children. Data from 10 children (5 males; 3.0–8.3 years, mean = 5.1) were collected simultaneously from MobilityLab and 3D motion capture during gait at selfselected and fast walking speeds. Spatiotemporal parameters were compared between the two methods using a Bland-Altman method. The results indicate that, while the temporal gait measurements were similar between the two systems, MobilityLab demonstrated a consistent bias with respect to measurement of the spatial data (stride length). This error is likely due to differences in relative leg length and gait characteristics in children compared to the MobilityLab adult reference population used to develop the stride length algorithm. A regression-based equation was developed based on the current data to correct the MobilityLab stride length output. The correction was based on leg length, stride time, and shank range-of-motion, each of which were independently associated with stride length. Once the correction was applied, all of the spatiotemporal parameters evaluated showed good agreement. The results of this study indicate that MobilityLab is a valid tool for gait analysis in typically-developing children. Further research is needed to determine the efficacy of this system for use in children suffering from pathologies that impact gait mechanics. ã 2016 Elsevier B.V. All rights reserved.
1. Introduction The past several years have seen increased use of inertial sensors to analyze movement in laboratory, clinic, and daily living environments [1]. By utilizing accelerometers, gyroscopes, or magnetometers (or a combination of these), inertial sensors can provide a wealth of data regarding the characteristics of global and segment-specific movement during a variety of tasks. Additionally, the sensors and recording equipment are relatively compact, portable, and low cost compared to traditional laboratory-based equipment (such as multi-camera 3D motion capture or instrumented mats), and can be used to collect human movement data in
* Corresponding author. E-mail address: [email protected] (J.L. Lanovaz). http://dx.doi.org/10.1016/j.gaitpost.2016.09.021 0966-6362/ã 2016 Elsevier B.V. All rights reserved.
environments and contexts where the use of traditional equipment is not possible. Inertial sensor technology that can be used in both laboratory and clinical environments has the potential to be a widely applicable method for researchers and clinicians to evaluate gait in a variety of healthy and clinical populations. One widely-used inertial sensor system is the MobilityLab system (APDM, Portland, OR). This system utilizes six inertial sensors, each containing tri-axial accelerometers, gyroscopes, and magnetometers providing a comprehensive evaluation of the spatiotemporal characteristics of motion during a variety of preprogrammed testing protocols [2–4]. Data collected from these sensors is transmitted wirelessly to a software program, which uses algorithms based on aggregated reference data that have been validated against both 3D motion capture and force plate data to calculate the spatiotemporal characteristics (such as stride time, stride length, and velocity of each stride) of movement [3–5]. The
J.L. Lanovaz et al. / Gait & Posture 51 (2017) 14–19
system is also capable of discriminating between different movements associated with various mobility tests such as the sit-to-stand and turning phases of the timed-up-and-go test [6], and has been used in the evaluation of gait and mobility in clinical populations including persons with Parkinson’s disease and multiple sclerosis [7,8]. While previous research has indicated that inertial sensor systems such as MobilityLab are a valid and reliable method of analyzing movement in adults [9], there has yet to be any research on the validity of their use in children. Since achieving functional gait and maximizing ambulatory independence are two of the most important functional outcomes for children suffering from musculoskeletal and neurological pathologies [10], it is crucial for clinicians to be able to analyze gait in children to recognize and attempt to correct any impairments and sub-optimal movement patterns that may be limiting functional capacity. Compared to traditional measurement tools used for gait analysis, inertial sensors offer several distinct benefits when working with children. The sensors are much easier to don and doff than reflective marker sets and use Velcro straps rather than adhesives, reducing the chances of skin irritation and/or discomfort during removal. Additionally, while most methods of gait analysis restrict movement to a given space or require the child to contact a target with their foot, the sensors allow the child to walk using their normal movement pattern with no environmental constraints. While some research exists evaluating the use of inertial sensors as a tool for gait analysis in children with cerebral palsy [11–13], there are currently no data evaluating the validity of inertial sensor systems relative to 3D motion capture (the gold standard of gait analysis). Direct measurements of kinematic parameters like linear acceleration and angular velocity from inertial systems are fairly accurate; however indirect measures such as spatiotemporal parameters often rely on algorithms with assumptions and reference values based on adult data. It is unclear if these approaches will result in accurate data when applied to children. The objective of this study is to validate the use of the MobilityLab inertial sensor system to obtain spatiotemporal parameters of gait in typically-developing children by comparing the level of agreement between data from the sensors and those obtained via 3D motion capture. We hypothesize that temporal data based on event detection will be accurate but estimations of spatial data may be influenced by adult-data assumptions inherent to the MobilityLab algorithms. 2. Methods Ten typically-developing children (five males) participated in the study (mean age 5.1 yrs, range 3.0 yrs–8.3 yrs). Participants were eligible for the study if they were between the ages of three and 10, free of any neurological disorders or lower limb musculoskeletal injuries, and were full term (37 weeks gestational age) at birth. The study was approved by the institutional Research Ethics Board and informed consent was obtained from the children’s guardians. In addition to obtaining informed consent from each child’s guardian, all of the participants gave verbal assent prior to their involvement in the study. Data were collected from each child as they walked in a straight line along a 7 m long walkway. Each child performed six to eight walking trials with approximately half the trials at a self-selected velocity. In the other half of the trials, the child was instructed to walk faster without running. The result was a range of walking velocities, with an overall mean velocity of 1.07 m/s, an average minimum of 0.83 m/s (SD 0.18 m/s) and an average maximum 1.51 m/s (SD 0.24 m/s). Encouragement was provided as needed to
15
maintain the child’s attention and engagement but the child walked without any hands-on assistance. Height was obtained to the nearest 0.1 cm using a stadiometer (mean 106.8 cm, range 93.5 cm–118.0 cm). Leg length (mean 42.7%, range 37.5 to 47.8%) was estimated as a percentage of total height from the motion capture data and was defined as the vertical length from the greater trochanter to the ankle (averaged across both legs) during standing. The MobilityLab system (version 1.0.0.201503302135) was used to collect the inertial-based spatiotemporal and kinematic data. Each child wore a total of six inertial sensors positioned on the dorsal side of both wrists, on the sternum close to the clavicular notch, on the lower back in correspondence to L4/L5, and on the frontal side of the shanks close to the malleoli. Data were collected wirelessly at a sampling rate of 128 Hz using MobilityLab’s iWalk module, which is designed for straight line walking of indeterminate length. Simultaneously, reference kinematic data were collected at 100 Hz using an 8 camera 3D motion capture system (Vicon Nexus, Centennial, CO). Reflective tracking markers (14 mm diameter) were fixed to the areas of the greater trochanter, lateral femoral condyle, lateral malleoli, heel and toe of both legs. Additionally, three tracking markers (9 mm diameter) were fixed to the inertial sensor on the sternum with two markers in line with a sensor axis and the third defining a cardinal plane in sensor coordinate system. Foot contact and lift off were detected from the motion capture data using a manually-tuned automated foot velocity threshold algorithm [14,15] and were used to calculate temporal data. Spatial data (i.e. stride length) were calculated using the heel position data. Inertial and motion capture data were synchronized during post-processing using custom software (Matlab R2006b, Mathworks, Natick, MA). Three dimensional acceleration of the sternum inertial sensor was calculated using motion capture data and transformed into the sensor coordinate system. These data were then time-matched to within 0.01 s of the corresponding raw sensor accelerometer data using a custom semi-automatic correlation method which used cross-correlation to provide an initial guess and then manual adjustment to find the final synchronization point. Data from six strides from each of the right and left legs were randomly chosen from each participant resulting in a total of 120 strides used for analysis. Four main spatiotemporal variables were compared between the systems; stride time (StrT), stance time (StnT), stride length (StrL) and stride velocity (StrV). StrT was defined as the time from heel strike on one foot to the next heel strike of the same foot. StrV was calculated on a stride-by-stride basis as the ratio of StrL and StrT. StnT was assessed in addition to StrT as both heel strike and toe off identification are included in StnT calculations. According to the MobilityLab manufacturer [16], the procedures used to generate temporal and stride length data from the inertial sensors are based on published algorithms [3,5]. For this study, the video-based motion capture system was assumed to be the gold standard. Data were compared between the two methods using the Bland-Altman method [17,18]. The BlandAltman method allows for comparisons between two different measurement systems to assess agreement when measuring the same set of data. The method provides an estimate of the bias between the systems and a measure of agreement known as limits of agreement (LoA). Since the magnitude of the some of the variables appeared to influence the bias, bias values and LoA were tested for significant non-zero slopes [18] and, when this was found, were fitted using linear regression [18]. Additionally, mean differences and root mean square (RMS) differences between the two systems were calculated. All statistics were calculated using SPSS (Version 23, IBM Corp, Armonk, NY).
16
J.L. Lanovaz et al. / Gait & Posture 51 (2017) 14–19
Fig. 1. Bland-Altman plots comparing MobilityLab (ML) and 3D motion capture (Vicon) results for (A) Stance time, (B) Stride time, (C) Stride length and (D) Stride velocity. Bias (solid line) and limits of agreement (dashed lines) are shown for each variable.
3. Results StnT and StrT were comparable between systems with little bias (Fig. 1). The LoA for StnT and StrT were approximately +/ 0.05 s and +/ 0.02 s respectively (Fig. 1). Even though the bias and LoA for the StrT were not constant, they were approximately equal to two motion capture sample point intervals. Examination of the raw foot fall data showed that 93% of initial contact and 68% of toe off identifications from the inertial system were within one frame of the motion capture system. StrL showed a large, non-constant bias ranging from +10%h to 10%h which indicated that MobilityLab overestimated the length of shorter strides and underestimated longer strides (Fig. 1). These differences would lead to 9–14% errors in StrL estimations at the
slowest and fastest walking velocities. StrV also showed a nonconstant bias with similar over and under estimation (Fig. 1) and a widening LoA indicating that agreement was worse at higher velocities. The errors in StrV were likely largely derived from StrL errors as the StrT differences were quite small. Mean and RMS differences are provided in Table 1. The error in StrL estimation for children using the MobilityLab algorithm was not unexpected. The StrL algorithm requires LL as well as the sagittal plane orientation of the shank and thigh at heel strike and toe off [5]. MobilityLab uses a value for LL that is a fixed percentage of total height [16] based on published data for adults (49.1%, [19] as referenced in [20]). LL in children is generally a smaller proportion of height as compared to adults [21] which was true in the current study (mean 42.7%). Additionally, the standard
Table 1 Mean and RMS differences between the two measurement systems. A positive value for the mean difference indicates that the MobilityLab values overestimate the motion capture data. Differences after the applied correction for StrL and StrV are also given. Original Data
StrT (s) StnT (s) StrL (%h) StrV (%h/s)
After Correction
Mean Difference
RMS Difference
0.001 s 0.003 s 4.68%h 4.97%h/s
0.014 s 0.026 s 10.30%h 12.10%h/s
%h indicates percent of the participant’s height.
Mean Difference
0.04%h 0.09%h
RMS Difference
6.43%h 7.80%h
J.L. Lanovaz et al. / Gait & Posture 51 (2017) 14–19
17
Table 2 Results for the step-wise linear regression modeling of the StrL error. The data from each step is given along with the subsequent R2 change. Model
Overall Adjusted R2
R2 change
Independent variables
Partial R2
b
Model 1 Model 2
0.22 0.35
n/a 0.14
Model 3
0.53
0.18
ShkROM ShkROM LL ShkROM LL StrT
0.22 0.23 0.18 0.30 0.30 0.29
0.48a 0.44a 0.37a 0.44a 0.45a 0.43a
a
Indicates significance at p < 0.001.
MobilityLab sensor configuration only collects angular motion data from the shank and uses a prediction formula, also based on adult data, to estimate the thigh position [3]; therefore, it was expected that the StrL error would be related to both LL and shank movement and would also likely be a function of walking speed, which would be reflected in StrT. We assessed the nature of the StrL error with bivariate correlations which showed that, on a stride-by-stride basis, StrL error was significantly correlated with shank range of motion (ShkROM) (r = 0.478, p < 0.001) and StrT (r = 0.334, p < 0.001) from MobilityLab. LL was also significantly correlated to average StrL error (r = 0.422, p < 0.001). ShkROM and StrT were considered surrogates for shank position and walking speed respectively. These values are reliable and are readily available from the standard MobilityLab output. ShkROM and StrT were not significantly correlated to each other while LL was minimally correlated with StrT (r = 0.184, p = 0.047) and was not correlated with ShkROM. A step-wise linear regression was used to model StrL error. Independent variables included ShkROM, StrT and LL. Results showed all three variables were significant in the model (Table 2). The corresponding regression formula (Eq. (1)) was then used to estimate StrL error for each stride: StrL error = 82.399
0.420 ShkROM
1.460 LL + 24.715 StrT(1)
where ShkROM is expressed in degrees, LL is in percent of total height, StrT is in seconds and StrL error is in percent of total height. The error correction was added to the original MobilityLab StrL to generate a corrected StrL value. The MobilityLab StrV was then recalculated using the corrected StrL. The corresponding BlandAltman plots showed greatly improved biases and a generally smaller LoA for StrV, although the StrL agreement tended to get worse at longer stride lengths (Fig. 2).
The mean StrL error was reduced from 4.7%h to 0.04%h and the RMS StrL error dropped from 10.3%h to 6.4%h after applying the correction. Similarly, the mean StrV error was reduced from 5.0%h/s to 0.1%h/s and the RMS StrV error dropped from 12.1%h/s to 7.8%h/s with the correction (Table 1). 4. Discussion This study examined the accuracy of MobilityLab, a commercial inertial sensor system, for analyzing basic spatiotemporal gait parameters in young children. MobilityLab produced accurate results for temporal gait variables which were comparable to what can be obtained with a 3D motion capture system. For the spatial variable of StrL, the MobilityLab algorithm overestimated short strides and underestimated long strides. A regression analysis was used to generate a StrL correction formula based on measured factors known to influence the MobilityLab algorithm. Improvements to MobilityLab StrL and StrV accuracy were seen with the application of the correction. The MobilityLab system was capable of detecting foot falls with a high degree of accuracy when compared to 3D motion capture. The LoA for StrT were approximately 2 motion capture frames intervals which is well within the normal error range of almost all foot fall detection algorithms [14,22–29]. The LoA for StnT were higher than for StrT which was likely a result of higher variability in detecting toe off events which is typical of automatic detection approaches [15]. The detection algorithm chosen for the 3D motion capture data in this study is just one of many available algorithms [15] and it is possible that a different approach would produce different values. Estimating StrL from a limited set of inertial sensors is challenging. The StrL algorithm employed in MobilityLab [5] was developed and tested for adults and, as such, has certain built
Fig. 2. Bland-Altman plots comparing corrected MobilityLab (ML) with 3D motion capture data (Vicon) for (A) Stride length and (B) Stride velocity. ML data were corrected using Eq. (1). Bias (solid lines) and limits of agreement (dashed lines) are shown.
18
J.L. Lanovaz et al. / Gait & Posture 51 (2017) 14–19
in assumptions which affect its performance when applied to children. The first assumption is that of fixed leg length proportionally with respect to height based on adult data. All of the children in this study had a LL that was a smaller proportion of their height than what is seen in a typical adult. Based on sample calculations we performed using the published algorithm [5], when LL is overestimated (i.e. using adult proportions rather than child), StrL will be consistently overestimated. This was the case for 70% of the StrL data in this study; however, since not all StrL were overestimated, other factors, such as limb kinematics, are also likely involved. MobilityLab is able to use a reduced set of sensors to analyze gait by utilizing an algorithm which estimates thigh angle based on shank data [3]. The underlying data structure for this estimation process was derived using data from adults walking on a treadmill [3]. Movement patterns in children can be significantly different than adults [30] and this likely affects the performance of the MobilityLab algorithm. The fact that ShkROM was one of the most significant predictors of StrL error in this study supports this claim. Stride velocity also appears to effect StrL estimation from MobilityLab as evidenced by the significant contribution of StrT as a StrL error predictor. There is likely a complex interaction between shank and thigh movement and walking velocity which would require specific calibration with children’s gait, similar to what was done in adults for the original algorithm [3]. The use of the correction equation presented in this paper significantly improved the StrL and StrV estimations from MobilityLab; mean StrL and StrV errors became negligible and absolute differences were reduced by an average of 40%. This indicates that, after the correction is applied, MobilityLab may be considered a valid substitute for a 3D motion-capture system at predicting the spatiotemporal parameters of gait in typicallydeveloping children. Additionally, since the factors included in the correction equation (ShkROM, StrT, and LL) are automatically generated from the MobilityLab data processing software, they are easy to access and enter into the correction equation after analysis. The limitations to this study included the low sample size, which makes it difficult to generalize the results of the regression. Since there were only 5 participants for each sex ranging from 3 to 10 years old, the descriptive data was highly variable. Additionally, body proportions and gait maturity develop and change throughout childhood and adolescence, which likely contributed to the substantial variation in descriptive parameters between participants [21]. The variables used in the regression were purposefully selected based on the known factors related to the calculation of StrL in the MobilityLab algorithm. The low sample size precluded a wide exploration of other possible factors but it is likely that the chosen variables are among the most influential. Although the potential for use of the MobilityLab system for use in children with pathological conditions affecting gait exists, further research is needed to determine if this system can accurately identify changes in spatiotemporal parameters in children with conditions/impairments that impact walking. 5. Conclusion Although previous studies have confirmed the validity of inertial sensors for use during gait analysis in adults, our study represents the first instance where the validity of inertial sensor measurements have been evaluated in children. Although we found the MobilityLab system to be a valid measure for temporal gait parameters and event detection, the system demonstrated a consistent bias with respect to StrL. A regression equation was used to correct for the observed bias, resulting in a substantial improvement in agreement between the MobilityLab and 3D motion capture systems with respect to StrL and StrV. The results
indicate that MobilityLab can be a valid tool for gait analysis in typically-developing children but that care must be taken when applying stride length algorithms developed for adult populations. Acknowledgements Funding for this study was provided by the Spina Bifida and Hydrocephalus Association of Canada. The sponsor was not directly involved in the design or the study, the collection/ analysis/interpretation of data, or the production of this manuscript. The authors would like to thank all of the participants. Conflict of interest The authors have no conflicts of interest. References [1] P.B. Shull, W. Jirattigalachote, M.A. Hunt, M.R. Cutkosky, S.L. Delp, Quantified self and human movement: a review on the clinical impact of wearable sensing and feedback for gait analysis and intervention, Gait Posture 40 (2014) 11–19. [2] A. Salarian, F.B. Horak, C. Zampieri, P. Carlson-Kuhta, J.G. Nutt, K. Aminian, iTUG, a sensitive and reliable measure of mobility, IEEE Trans. Neural Syst. Rehabil. Eng. 18 (2010) 303–310. [3] A. Salarian, P.R. Burkhard, F.J. Vingerhoets, B.M. Jolles, K. Aminian, A novel approach to reducing number of sensing units for wearable gait analysis systems, IEEE Trans. Biomed. Eng. 60 (2013) 72–77. [4] M. Mancini, L. King, A. Salarian, L. Holmstrom, J. McNames, F.B. Horak, Mobility lab to assess balance and gait with synchronized body-worn sensors, J. Bioeng. Biomed. Sci. (Suppl. (1)) (2011) 007. [5] K. Aminian, B. Najafi, C. Büla, P. Leyvraz, P. Robert, Spatio-temporal parameters of gait measured by an ambulatory system using miniature gyroscopes, J. Biomech. 35 (2002) 689–699. [6] J.T. Coulthard, T.T. Treen, A.R. Oates, J.L. Lanovaz, Evaluation of an inertial sensor system for analysis of timed-up-and-go under dual-task demands, Gait Posture 41 (2015) 882–887. [7] C. Zampieri, A. Salarian, P. Carlson-Kuhta, K. Aminian, J.G. Nutt, F.B. Horak, The instrumented timed up and go test: potential outcome measure for disease modifying therapies in Parkinson's disease, J. Neurol. Neurosurg. Psychiatry 81 (2010) 171–176. [8] R.I. Spain, R.J. St George, A. Salarian, M. Mancini, J.M. Wagner, F.B. Horak, D. Bourdette, Body-worn motion sensors detect balance and gait deficits in people with multiple sclerosis who have normal walking speed, Gait Posture 35 (2012) 573–578. [9] A. Muro-de-la-Herran, B. Garcia-Zapirain, A. Mendez-Zorrilla, Gait analysis methods: an overview of wearable and non-wearable systems, highlighting clinical applications, Sensors 14 (2014) 3362–3394. [10] S.K. Campbell, R.J. Palisano, D.W. Vander Linden, Physical Therapy for Children, Elsevier Saunders, 2006. [11] A.G. Cutti, A. Ferrari, P. Garofalo, M. Raggi, A. Cappello, A. Ferrari, ‘Outwalk’: a protocol for clinical gait analysis based on inertial and magnetic sensors, Med. Biol. Eng. Comput. 48 (2010) 17–25. [12] C. van den Noort Josien, A. Ferrari, A.G. Cutti, J.G. Becher, J. Harlaar, Gait analysis in children with cerebral palsy via inertial and magnetic sensors, Med. Biol. Eng. Comput. 51 (2013) 377–386. [13] A.B. Bourgeois, B. Mariani, K. Aminian, P. Zambelli, C. Newman, Spatiotemporal gait analysis in children with cerebral palsy using, foot-worn inertial sensors, Gait Posture 39 (2014) 436–442. [14] S. Ghoussayni, C. Stevens, S. Durham, D. Ewins, Assessment and validation of a simple automated method for the detection of gait events and intervals, Gait Posture 20 (2004) 266–272. [15] D.A. Bruening, S.T. Ridge, Automated event detection algorithms in pathological gait, Gait Posture 39 (2014) 472–477. [16] M. Johnson, APDM Inc., Personal communication. [17] J.M. Bland, D. Altman, Statistical methods for assessing agreement between two methods of clinical measurement, Lancet 327 (1986) 307–310. [18] J.M. Bland, D.G. Altman, Measuring agreement in method comparison studies, Stat. Methods Med. Res. 8 (1999) 135–160. [19] R. Drillis, R. Contini, Body Segment Parameters Report No 1163-03 Office of Vocational Rehabilitation, Department of health, Education and Welfare, New York, 1966. [20] D.A. Winter, Biomechanics and Motor Control of Human Movement, John Wiley & Sons, 2009. [21] B. Bogin, M.I. Varela-Silva, Leg length, body proportion, and health: a review with a note on beauty, Int. J. Environ. Res. Public Health 7 (2010) 1047–1075. [22] A. De Asha, M. Robinson, G. Barton, A marker based kinematic method of identifying initial contact during gait suitable for use in real-time visual feedback applications, Gait Posture 36 (2012) 650–652.
J.L. Lanovaz et al. / Gait & Posture 51 (2017) 14–19 [23] E. Desailly, Y. Daniel, P. Sardain, P. Lacouture, Foot contact event detection using kinematic data in cerebral palsy children and normal adults gait, Gait Posture 29 (2009) 76–80. [24] A. Hreljac, R.N. Marshall, Algorithms to determine event timing during normal walking using kinematic data, J. Biomech. 33 (2000) 783–786. [25] B. Hsue, F. Miller, F. Su, J. Henley, C. Church, Gait timing event determination using kinematic data for the toe walking children with cerebreal palsy, J. Biomech. 40 (2007) S529. [26] J.M. Jasiewicz, J.H. Allum, J.W. Middleton, A. Barriskill, P. Condie, B. Purcell, R.C. T. Li, Gait event detection using linear accelerometers or angular velocity transducers in able-bodied and spinal-cord injured individuals, Gait Posture 24 (2006) 502–509.
19
[27] C.M. O’Connor, S.K. Thorpe, M.J. O’Malley, C.L. Vaughan, Automatic detection of gait events using kinematic data, Gait Posture 25 (2007) 469–474. [28] J. Salazar-Torres, Validity of an automated gait event detection algorithm in children with cerebral palsy and non-impaired children, Gait Posture 24 (2006) S130–1. [29] J. Zeni, J. Richards, J. Higginson, Two simple methods for determining gait events during treadmill and overground walking using kinematic data, Gait Posture 27 (2008) 710–714. [30] A.W. Froehle, R.W. Nahhas, R.J. Sherwood, D.L. Duren, Age-related changes in spatiotemporal characteristics of gait accompany ongoing lower limb linear growth in late childhood and early adolescence, Gait Posture 38 (2013) 14–19.