Towards Mobile Gait Analysis: Concurrent Validity and Test-Retest Reliability of an Inertial Measurement System for the Assessment of Spatio-Temporal Gait Parameters

Felix Kluge, Heiko Gassner, Julius Hannink, Cristian F. Pasluosta, Jochen Klucken, Björn M. Eskofier

TL;DR

This paper introduces and validates a novel wearable inertial measurement unit (IMU) system for gait analysis, demonstrating its concurrent validity against a gold-standard motion capture system and excellent test-retest reliability. The system accurately captures spatio-temporal gait parameters, making it a promising tool for objective, long-term monitoring of gait in both healthy individuals and those with Parkinson's disease.

What Problem Does This Paper Address?

Clinically assessing gait impairments, particularly in neurodegenerative diseases like Parkinson's disease (PD), traditionally relies on subjective observational ratings. While gold-standard motion capture systems offer objective data, they are expensive, stationary, and require manual post-processing, limiting their feasibility in routine clinical practice and long-term monitoring. Existing wearable systems often lack comprehensive validation for both concurrent validity against reference systems and test-retest reliability, especially in patient populations. This study aims to bridge this gap by developing and rigorously evaluating a mobile, sensor-based gait analysis system that is both accurate and reliable for clinical applications.

Methods

The study employed a novel sensor-based gait analysis system utilizing two Shimmer3 IMUs attached to the ankles of participants' shoes. These sensors captured tri-axial gyroscope and accelerometer data at 102.4 Hz. The data was wirelessly transmitted to a mobile device for storage.

Key methodological aspects include:

  • Sensor Calibration: A calibration procedure was used to convert raw sensor signals into physical units, with specific attention paid to aligning sensor axes due to their mirrored mounting.
  • Stride Segmentation: A multi-dimensional subsequence dynamic time warping (msDTW) approach, previously shown to be effective for various populations, was used to detect individual strides from the continuous inertial data stream.
  • Gait Event Detection: Heel strike (HS), toe off (TO), and mid-stance (MS) events were identified from the inertial data.
  • Spatio-Temporal Parameter Calculation:
    • Temporal Parameters: Stride time, stance time, and swing time were directly calculated from the detected gait events.
  • Spatial Parameters: Foot orientation was estimated using the Euston complementary filter, which combines gyroscope and accelerometer data. Gravity was removed from the transformed acceleration signal. Foot trajectory was then derived through a double integration algorithm that incorporated drift correction techniques to mitigate integration errors. Stride length and gait velocity were calculated from these trajectories.
  • Reference System: A markerless optical motion capture system (Simi Reality Motion Systems) with eight cameras served as the gold standard for comparison.
  • Statistical Analysis:
    • Concurrent Validity: Assessed using Pearson's correlation, bias (mean difference), absolute error, and relative absolute error between the sensor system and the motion capture system. Bland-Altman plots were used to visualize agreement and identify systematic errors.
  • Test-Retest Reliability: Evaluated using intraclass correlation coefficients (ICC) for five healthy subjects over three measurement sessions within a two-week period. Both single-measurement (ICC(2,1)) and average-measurement (ICC(2,k)) reliabilities were calculated.

Datasets

The study utilized data collected from two groups:

  • Healthy Subjects: Eleven healthy individuals.
  • Parkinson's Disease (PD) Patients: Four patients diagnosed with PD.

Data was collected during controlled walking tasks at three different self-selected speeds (normal, slow, fast) over a 10-meter walkway. A total of 1166 strides were analyzed for concurrent validity. For test-retest reliability, data from five healthy subjects (10 legs) across three sessions were used.

Key Insights and Ideas

  • Bridging the Gap: The study successfully demonstrates that a wearable IMU system can achieve concurrent validity comparable to a gold-standard motion capture system for spatio-temporal gait parameters. This is crucial for moving gait analysis out of the lab and into real-world settings.
  • Robust Algorithms: The combination of msDTW for stride segmentation, the Euston complementary filter for orientation, and a drift-correction algorithm for trajectory estimation proved effective in extracting accurate gait parameters from inertial sensor data.
  • PD Patient Applicability: The system showed no significant difference in accuracy between healthy subjects and PD patients, suggesting its potential for assessing gait impairments in this population.
  • Temporal Discretization Impact: The temporal resolution of the camera system (100 Hz) and sensor system (102.4 Hz) introduced slight limitations in the accuracy of temporal measurements, leading to regular patterns in stance and swing time differences.
  • Velocity as a Potential Challenge: While overall reliability was high, gait velocity showed slightly higher errors at higher walking speeds and consistently lower ICC values, suggesting it might be a more unstable parameter to estimate accurately with this system.

Key Results

  • Concurrent Validity:
    • High correlations (r > 0.93) were observed for all gait parameters between the sensor system and the motion capture system.
  • Bias for stride time was negligible (0 ± 16 ms), and for stride length was small (1.4 ± 6.7 cm).
  • Absolute relative errors were below 9% for all parameters, with spatial parameters (stride length and velocity) showing errors less than 4%.
  • Agreement was similar between healthy subjects and PD patients, with no systematic magnitude-dependent errors observed.
  • Test-Retest Reliability:
    • Excellent reliability (ICC > 0.81) was found for stride time, stance time, swing time, and stride length.
  • Gait velocity showed good reliability in fast and normal walking conditions (ICC > 0.75) but fair reliability in slow walking (ICC = 0.55).
  • When considering the average of three measurements (ICC(2,k)), all parameters achieved excellent reliability above 0.79.

Limitations and Future Work

  • Limited PD Sample Size: The study included only four PD patients, which is a small sample size. The authors acknowledge that for severely impaired gait, stride segmentation might be less reliable, and future work should evaluate the system's performance with more diverse and severely affected patient groups.
  • Laboratory Environment: The study was conducted in a controlled laboratory setting. The authors suggest that future research should assess the system's validity in unsupervised, free-living environments, as laboratory conditions might influence gait patterns and affect external validity.
  • Straight Walking Only: Only straight walking was evaluated. Future studies should investigate the system's accuracy during turning movements and other complex gait scenarios.
  • Automatic Synchronization: While synchronization between the camera and sensor systems was performed, the authors recommend incorporating automatic synchronization between the two foot sensors to enable accurate calculation of parameters like double limb support.
  • Velocity Instability: The consistently lower ICCs for gait velocity suggest it may be an unstable parameter. Further investigation into why velocity is less reliable is warranted.
  • Healthy Subject Focus for Reliability: Test-retest reliability was only assessed in healthy adults. The authors note that evaluating reliability in clinical populations is challenging due to potential symptom variability but is crucial for future applications.

Summary updated

BibTeX
@inproceedings{kluge2017towards,
  author = {Felix Kluge and Heiko Gassner and Julius Hannink and Cristian F. Pasluosta and Jochen Klucken and Björn M. Eskofier},
  title = {{Towards Mobile Gait Analysis: Concurrent Validity and Test-Retest Reliability of an Inertial Measurement System for the Assessment of Spatio-Temporal Gait Parameters}},
  booktitle = {Sensors},
  year = {2017},
  url = {https://openreview.net/forum?id=21R53dpQXv},
}
Figure