Assessing Physical Rehabilitation Exercises using Graph Convolutional Network with Self-supervised regularization.

Area of Science:

• Computer Vision
• Rehabilitation Medicine
• Machine Learning

Background:

• Computer-vision offers low-cost, portable, real-time physical rehabilitation exercise evaluation.
• Current deep learning methods, like convolutional neural networks (CNNs), struggle with spatial human body information, limiting assessment accuracy.
• Automatic assessment of physical rehabilitation exercises is crucial for effective patient recovery.

Purpose of the Study:

• To develop a more accurate deep learning framework for assessing physical rehabilitation exercises.
• To overcome the limitations of existing CNN-based methods by incorporating spatial body information.
• To enhance the accuracy of automated physical rehabilitation assessments through self-supervised learning.

Main Methods:

• Proposed a deep learning framework utilizing a graph convolutional network (GCN).
• Incorporated self-supervised regularization to improve prediction accuracy.
• Utilized 3D body-landmark sequences for exercise analysis.

Main Results:

• The proposed GCN method achieved state-of-the-art performance on a benchmark dataset.
• Demonstrated lower error rates compared to existing CNN methods.
• Self-supervised learning significantly improved prediction accuracy in exercise assessments.

Conclusions:

• The developed GCN framework provides a more accurate method for assessing physical rehabilitation exercises.
• This computer-vision-based approach enables automated, real-time evaluations in home environments.
• The system offers potential for remote patient feedback, reducing the need for in-person supervision.