Evaluation of Deep Learning-Based Event Detection for Parameter Estimation During Complex Walking in Parkinson's...

Area of Science:

• Biomechanics
• Wearable Technology
• Deep Learning

Background:

• Quantifying complex walking tasks beyond steady-state gait (SSG) remains challenging despite advances in wearable sensors.
• There is a need for reliable methods to analyze gait in real-world environments.

Purpose of the Study:

• To evaluate an inertial sensor-based processing pipeline for quantifying complex walking tasks.
• To assess a deep learning method for event detection during stride segmentation and subsequent gait parameter calculation.

Main Methods:

• Utilized a Temporal Convolutional Network (TCN) for stride segmentation.
• Employed established methods for trajectory reconstruction and parameter extraction.
• Validated against a pressure walkway as the reference system for steady state gait (SSG), turns, gait initiation (GI), and gait termination (GT) strides.

Main Results:

• Achieved small mean errors (≤ 1 ms temporal, ≤ 2.3 cm spatial) and strong correlation (r ≥ 0.96) for SSG strides.
• Demonstrated similar high performance for turn, GI, and GT strides (≤ 7 ms temporal, ≤ 2.9 cm spatial; r ≥ 0.95).

Conclusions:

• Inertial measurement unit (IMU) derived gait metrics using TCN event detection and Gaitmap functions reliably quantify gait during simple and complex walking tasks.
• The proposed method offers a robust approach for understanding mobility in diverse environments.