Validation of a Sensor-Based Gait Analysis System with a Gold-Standard Motion Capture System in Patients with Parkinson's Disease

Verena Jakob, Arne Küderle, Felix Kluge, Jochen Klucken, Bjoern M. Eskofier, Jürgen Winkler, Martin Winterholler, Heiko Gassner2021 · DOI

TL;DR

This study validates a wearable sensor-based gait analysis system against a gold-standard motion capture system in Parkinson's Disease (PD) patients. The findings demonstrate that the sensor system accurately captures key gait parameters related to bradykinesia and short steps, making it a promising tool for remote and long-term monitoring of PD.

What Problem Does This Paper Address?

Parkinson's Disease (PD) is a progressive neurological disorder that significantly impacts gait, leading to an increased risk of falls and reduced quality of life. Accurately assessing gait is crucial for diagnosis, monitoring disease progression, and evaluating treatment effectiveness. While 3D motion capture (MC) systems are considered the gold standard for gait analysis, they are expensive, time-consuming, and confined to laboratory settings, limiting their use in real-world clinical applications and home environments. Wearable sensor-based systems offer a more accessible and practical alternative for continuous gait monitoring. However, before these systems can be widely adopted, their technical validity and reliability must be rigorously established, especially in specific patient populations like those with PD, who exhibit unique gait abnormalities. This paper aims to bridge this gap by validating a specific wearable sensor system against MC in PD patients.

Methods

The researchers employed a two-pronged approach to validate their mobile gait analysis (MGL) system. First, they conducted a reliability analysis comparing gait parameters measured by the MGL with those from a gold-standard 3D motion capture (MC) system. This involved having 37 PD patients and 14 healthy controls perform a standardized 5x10m walking test. Gait parameters were extracted from both systems, and their agreement was assessed using intra-class correlation coefficients (ICCs) and Bland-Altman plots. ICCs measure the consistency of measurements, while Bland-Altman plots visualize the agreement between two methods by showing the bias and limits of agreement.

Second, they investigated the clinical relevance of the MGL by comparing gait parameters between the PD cohort and the healthy control group. This was done to determine if the sensor system could effectively distinguish between individuals with and without PD, and if it could capture the characteristic gait impairments associated with PD. Statistical analysis included the Kolmogorov-Smirnov test for normality, Levene test for variance homogeneity, Wilcoxon rank test for comparing systems, and the Mann-Whitney U test for comparing groups. To account for multiple comparisons, the Benjamini-Hochberg False-Rate-Discovery test was applied.

Datasets

The study utilized data collected from 37 patients diagnosed with sporadic Parkinson's Disease and 14 age-matched healthy control participants. No new benchmark datasets were introduced.

Key Insights and Ideas

  • Wearable sensors can reliably capture key PD gait features: The study strongly suggests that wearable sensors can accurately measure gait parameters that are critical for understanding bradykinesia and short steps, which are hallmark symptoms of PD.
  • Shuffling gait parameters require careful interpretation: While most parameters showed good agreement, shuffling gait parameters (heel strike, toe off, and toe clearance) exhibited lower reliability and significant differences between the sensor and MC systems, suggesting potential differences in how these specific gait phases are detected.
  • Sensor systems can differentiate PD from healthy controls: The MGL system, like the MC system, was able to distinguish between PD patients and healthy controls, indicating its potential for clinical use in identifying gait abnormalities.
  • Early PD gait differences are subtle: The study found that both systems struggled to differentiate between early-stage PD patients (H&Y stages 1+2) and healthy controls, suggesting that more advanced gait impairments might be necessary for clear detection by these methods. This aligns with findings that gait initiation and turning are more sensitive indicators in early PD.
  • The limitations of lab-based MC pave the way for wearables: The inherent limitations of MC (cost, time, lab-bound nature) highlight the significant advantage of wearable sensors for continuous, real-world gait monitoring, which is crucial for managing chronic conditions like PD.

Key Results

  • Excellent reliability for bradykinesia and short steps: Gait parameters reflecting bradykinesia and short steps (gait speed, stride length, stride time, relative stance time, and relative swing time) showed excellent reliability with ICC values > 0.90.
  • Good reliability for shuffling gait: Shuffling gait parameters (heel strike (HS), toe off (TO), and toe clearance (TC)) demonstrated good reliability with ICCs > 0.822.
  • No significant differences in most parameters between systems: For bradykinesia and short steps, most parameters showed no significant difference between the MGL and MC systems. Stride time was the only exception, showing a significant difference (p = 0.003).
  • Significant differences in shuffling gait parameters: Shuffling gait parameters (HS, TO, and TC) showed significant differences between the MGL and MC systems (p < 0.001).
  • Both systems differentiate PD from controls: Both MGL and MC systems successfully identified significant differences in gait parameters between PD patients and healthy controls.
  • Advanced PD stages are more clearly differentiated: Significant differences were observed between PD patients with H&Y stage 3 and controls. However, no significant differences were detected between patients with H&Y stages 1+2 and controls.

Limitations and Future Work

The authors acknowledge several limitations:

  • Limited measurement volume: The camera setup for the MC system restricted the measurement volume to a 3x6 meter area, meaning only the middle part of the gait cycle could be compared. This excluded gait variability parameters and may not fully represent daily life scenarios.
  • Synchronization method: The stride-by-stride synchronization was implemented via video image by an investigator and not fully automated, which could introduce some subjectivity.
  • Sample size and distribution: The differing sample sizes between the PD and control groups, as well as the diverse distribution of H&Y stages within the PD group, might influence the results.

Future work should focus on:

  • Investigating shuffling gait parameter discrepancies: Further research is needed to clarify the reasons for the differences in shuffling gait parameters between the MGL and MC systems, potentially exploring different gait phase determination algorithms.
  • Exploring gait initiation and turning: The study acknowledges that gait initiation, termination, and turning are more sensitive indicators of motor impairment in early PD and suggests future studies should investigate these aspects.
  • Evaluating therapy response and disease progression: The authors propose that future studies should investigate whether sensor-based gait parameters can objectively reflect therapy response and disease progression in PD.
  • Expanding to other neurological conditions: The potential of wearable sensors for evaluating disease evolution and therapy compliance in other orthopedic, neuromuscular, and neurological diseases should be explored.

Summary updated

BibTeX
@misc{jakob2021validation,
  author = {Verena Jakob and Arne Küderle and Felix Kluge and Jochen Klucken and Bjoern M. Eskofier and Jürgen Winkler and Martin Winterholler and Heiko Gassner},
  title = {{Validation of a Sensor-Based Gait Analysis System with a Gold-Standard Motion Capture System in Patients with Parkinson's Disease}},
  year = {2021},
  doi = {10.3390/s21227680},
}
Figure