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

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Artificial Intelligence in Healthcare

Background:

• Gait rehabilitation in post-stroke hemiplegia is vital for mobility and quality of life.
• Traditional gait analysis relies on manual feature extraction, leading to inaccuracies and bias.
• Understanding individual gait patterns is crucial for effective rehabilitation strategies.

Purpose of the Study:

• To develop and validate an end-to-end deep learning approach for autonomous gait pattern clustering in post-stroke hemiplegic patients.
• To minimize human intervention in gait feature extraction and analysis.
• To identify distinct gait clusters using kinematic data from joint angle trajectories.

Main Methods:

• A cross-sectional study involving 74 sub-acute post-stroke hemiplegic patients.
• Utilized deep temporal clustering on sagittal plane joint angle and angular velocity trajectories (hip, knee, ankle) during gait cycles.
• Employed end-to-end optimization for simultaneous feature extraction and clustering, with tailored hyperparameter tuning.

Main Results:

• Identified six optimal gait clusters with a silhouette score of 0.2831, outperforming other clustering algorithms.
• Demonstrated the effectiveness of the deep learning approach in clustering gait patterns without manual feature engineering.
• Presented detailed statistical analysis of spatiotemporal, kinematic, and clinical features for each identified cluster.

Conclusions:

• End-to-end deep learning offers a significant performance improvement for gait pattern analysis in post-stroke hemiplegia.
• This automated approach reduces bias and enhances accuracy compared to manual feature extraction methods.
• Future research should incorporate multi-planar data and other biomechanical factors for a more holistic gait analysis.