JMIR Publications

ABSTRACT

Background:

Gait impairment constitutes a prevalent stroke sequela. While observational gait analysis remains a standard clinical practice for assessing neuromotor impairments in people with stroke, it is prone to subjective bias. Consequently, objective assessment is required to inform individualized rehabilitation protocols.

Objective:

This study aims to validate a deep learning framework for automating clinical gait assessment using kinematic data.

Methods:

This study was conducted in a university-affiliated gait analysis laboratory. Kinematic data from 51 hemiparetic individuals and 18 healthy controls was collected using wearable inertial measurement units (IMUs) and video recordings. Wisconsin Gait Scale (WGS) rating scores assessed by expert physiotherapists served as the reference standard. A Statistical Graph Convolutional Network (STAT-GCN), incorporating a customized, interpretable STAT Attention Head (SA Head) module, was designed to automate WGS prediction from kinematic data. The SA Head employs a contrastive learning strategy to enhance discrimination between highly confusable samples.

Results:

Performance was benchmarked against four final-year physiotherapy students. STAT-GCN achieved >78% accuracy across all 13 WGS items, outperforming all final-year physiotherapy students by an average of 18%.

Conclusions:

The proposed STAT-GCN framework provides an automated, clinically viable solution for objective gait assessment in stroke rehabilitation. Notably, it demonstrates superior diagnostic accuracy compared to novice clinicians, while its interpretable architecture supports trustworthy clinical decision-making. By providing a standardized and interpretable scoring method, it mitigates subjective bias in manual assessment and significantly reduces the administrative workload for physiotherapists.