Unified Gait Event Detection using Temporal Convolutional Network and Bayesian Optimization.

Area of Science:

• Biomechanics and Biomedical Engineering
• Machine Learning and Artificial Intelligence
• Signal Processing

Background:

• Accurate gait event detection (GED) using inertial measurement unit (IMU) signals is crucial for analyzing gait abnormalities.
• Existing GED methods often lack generalizability, requiring activity-specific heuristics or multiple deep learning models for different environments.
• This limits scalability and real-world applicability in clinical settings.

Purpose of the Study:

• To develop a unified deep learning framework for robust gait event detection (GED) across diverse activities and environments.
• To overcome the limitations of existing methods by eliminating the need for separate models for each condition.
• To reduce computational costs and enhance the generalizability of GED.

Main Methods:

• A temporal convolutional network (TCN) forms the core of the unified deep learning framework.
• Bayesian optimization was employed for efficient TCN hyperparameter tuning.
• Gaussian kernel-based ground truth generation and a weighted loss function were utilized to enhance performance, particularly for challenging gait events.

Main Results:

• The unified framework achieved high accuracy in gait event detection, with F1 scores of 0.99 ± 0.00 for heel strikes (HS) and 0.97 ± 0.06 for toe-offs (TO).
• Competitive temporal precision was demonstrated, indicated by a low mean absolute error (MAE).
• Performance surpassed existing state-of-the-art methods in leave-one-out cross-validation across 20 subjects.

Conclusions:

• The proposed unified deep learning framework offers a scalable and generalizable solution for gait event detection (GED).
• It demonstrates significant potential for robust real-world applications in clinical gait analysis, rehabilitation, and assistive technology development.
• The framework's ability to operate across multiple activities and environments without model retraining represents a substantial advancement.