Temporal Convolutional Network for Gait Event Detection.

Area of Science:

• Biomedical Engineering
• Machine Learning
• Human Movement Analysis

Background:

• Accurate gait event detection (GED) is crucial for identifying biomechanical markers of movement disorders.
• Existing GED methods struggle with accuracy in complex, real-world walking scenarios.
• Challenges include variability in daily activities and diverse environmental conditions.

Purpose of the Study:

• To develop a robust deep learning framework for automatic GED in complex walking scenarios.
• To enhance the accuracy and reliability of identifying key gait events like Heel Strike (HS) and Toe-Off (TO).
• To evaluate the framework's performance across diverse indoor and outdoor activities.

Main Methods:

• A novel framework utilizing a Temporal Convolution Network (TCN) for gait event detection.
• Implementation of a peak detection algorithm as a post-processing step for precise event identification.
• Evaluation using a public gait dataset, measuring performance with F1 score and Mean Absolute Error (MAE).

Main Results:

• The framework achieved a mean F1-score of 0.96 ± 0.07 for HS and 0.92 ± 0.11 for TO.
• Mean Absolute Error (MAE) for time agreement was 6.25 ms ± 3.67 ms for HS and 16.87 ms ± 11.56 ms for TO.
• Consistent performance was observed across various indoor and outdoor walking conditions.

Conclusions:

• The proposed deep learning framework demonstrates robust and accurate gait event detection in diverse conditions.
• The methodology shows significant potential for applications in analyzing pathological gaits during daily life.
• This approach offers a reliable tool for biomechanical analysis and movement disorder assessment.