Toward Personalized Rotator Cuff Physical Therapy Dosage Using a Machine Learning-Based Pilot Study with EMG.

Area of Science:

• Biomedical Engineering
• Rehabilitation Science
• Machine Learning in Medicine

Background:

• Rotator cuff injuries are prevalent, impacting shoulder function and quality of life.
• Current rehabilitation protocols lack personalization, posing a clinical challenge.
• Optimizing exercise dosage is crucial for effective recovery.

Purpose of the Study:

• To develop a machine learning approach for personalized rotator cuff rehabilitation.
• To utilize surface electromyography (sEMG) data to model muscle activation patterns.
• To optimize therapy session timing based on muscle activation goals.

Main Methods:

• Collected sEMG data from eight healthy individuals during four key shoulder movements.
• Employed the XGBoost algorithm to model muscle activation patterns.
• Utilized a linear programming model to allocate therapy time, optimizing for superficial and deep muscle activation.

Main Results:

• XGBoost achieved high predictive accuracy (R² = 0.5325) in modeling muscle activation.
• Optimized time allocations favored external rotation at 90° abduction and scaption.
• Three scenarios demonstrated varying emphasis on superficial muscle activation and deep muscle strain minimization.

Conclusions:

• Machine learning can personalize rotator cuff rehabilitation, improving treatment plans.
• Data-driven approaches enhance clinical decision-making in complex muscle activation analysis.
• This method offers adaptive planning with strong performance and low latency for clinical application.