Machine-learning models for shoulder rehabilitation exercises classification using a wearable system.

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Machine Learning in Healthcare

Background:

• Shoulder rehabilitation is crucial for patients with rotator cuff tears.
• Accurate exercise classification is essential for effective rehabilitation and monitoring.
• Current methods may lack objective, real-time feedback.

Purpose of the Study:

• To train and evaluate machine learning models for automated classification of shoulder rehabilitation exercises.
• To assess the feasibility of using wearable sensors for this task.

Main Methods:

• Trained six supervised machine learning models (including Random Forest) using data from magneto-inertial sensors.
• Utilized data from 19 healthy subjects and 17 patients with rotator cuff tears performing six specific exercises.
• Evaluated classification performance using nested cross-validation.

Main Results:

• The Random Forest classifier achieved the highest performance, with 89.91% accuracy and an 89.89% F1-score.
• Demonstrated high accuracy in distinguishing between different shoulder rehabilitation exercises.

Conclusions:

• Wearable sensors combined with machine learning effectively classify shoulder rehabilitation exercises.
• The system shows promise for remote, home-based monitoring, reducing patient burden.
• The proposed system is feasible, effective, and user-friendly for patient-driven sensor placement.