An Expert-Knowledge-Based Graph Convolutional Network for Skeleton- Based Physical Rehabilitation Exercises...

Area of Science:

• Rehabilitation Medicine
• Artificial Intelligence
• Biomedical Engineering

Background:

• Physical therapists face challenges in providing individualized guidance due to therapist-patient ratios.
• Automated assessment of rehabilitation exercises is needed for accurate movement quantification and feedback.
• Current methods struggle with precise evaluation of patient training quality.

Purpose of the Study:

• To develop an automated system for assessing the quality of physical rehabilitation exercises.
• To enhance the accuracy and interpretability of automated rehabilitation assessments.
• To provide therapists with tools to better understand patient performance and guide recovery.

Main Methods:

• An Expert-knowledge-based Graph Convolutional approach was developed for automated exercise assessment.
• Utilized expert knowledge to enhance spatial feature extraction in Graph Convolutional modules.
• Incorporated a Gated pooling module for feature aggregation and a Transformer module for temporal dependency analysis.

Main Results:

• The proposed method achieved state-of-the-art performance on the KIMORE dataset.
• Demonstrated significant improvements in accurately quantifying rehabilitation movements.
• Attention scores and weight matrices provided interpretable insights into spatial and temporal movement dimensions.

Conclusions:

• The Expert-knowledge-based Graph Convolutional approach effectively automates physical rehabilitation exercise assessment.
• The system offers enhanced accuracy and interpretability, aiding therapists and patients.
• This method represents a significant advancement in automated physical therapy and recovery monitoring.