Deep Temporal Clustering of Pathological Gait Recovery Patterns in Post-Stroke Patients Using Joint-Angle...

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Science

Background:

• Stroke survivors with hemiplegia often exhibit long-term gait impairments.
• Understanding longitudinal gait recovery patterns is crucial for effective rehabilitation strategies.
• Existing methods struggle with the complexity and data scarcity of long-term gait analysis.

Purpose of the Study:

• To analyze long-term gait recovery patterns in sub-acute post-stroke hemiplegic patients.
• To apply end-to-end deep learning (DL)-based clustering to sagittal joint-angle trajectories.
• To address data scarcity in long-term gait trajectory datasets using data augmentation.

Main Methods:

• Employed TimeVAE and Diffusion-TS for time-series data augmentation to generate synthetic gait trajectories.
• Utilized a Deep Temporal Clustering (DTC) model for joint learning of temporal representations and cluster assignments.
• Applied clustering evaluation criteria to determine the optimal number of patient recovery groups.

Main Results:

• Identified six distinct clusters representing individualized longitudinal gait recovery patterns.
• These clusters were characterized by statistically significant and unique kinematic features.
• Demonstrated the effectiveness of DL-based clustering in capturing complex recovery dynamics.

Conclusions:

• This study pioneers the use of deep clustering for analyzing longitudinal gait recovery in post-stroke patients.
• The findings provide a conceptual framework for future research in stroke rehabilitation and gait analysis.
• The developed methodology offers a novel approach to understanding patient-specific recovery trajectories.