IMU-based rotator cuff injury recognition with varying configurations and combinations

Area of Science:

• Biomechanics and Movement Science
• Medical Artificial Intelligence
• Orthopedics and Sports Medicine

Background:

• Functional movement assessment is crucial for evaluating shoulder impairments and daily activity capacity.
• Rotator cuff injury (RCI) recognition aids clinical management, but accurate and cost-effective methods are needed.

Purpose of the Study:

• To develop accurate, cost-effective RCI recognition models using machine learning (ML)/deep learning (DL) with upper limb kinematic data.
• To identify optimal motion task combinations and model configurations for clinical application.

Main Methods:

• 102 participants (51 RCI, 25 adhesive capsulitis, 26 healthy) performed six functional movements and two range of motion (ROM) tests.
• Upper limb kinematic data collected via IMUs were analyzed using eight ML/DL models, including the proposed RCIRNet.
• Data augmentation was used due to the small sample size; performance evaluated via Accuracy, Precision, Recall, F1-score, and AUC.

Main Results:

• The shoulder frontal ROM test alone with RCIRNet achieved high performance (Accuracy 0.89, F1-score 0.89, AUC 0.93).
• Combining four tasks improved Recall to 1.0 while maintaining high Accuracy, F1-score, and AUC.
• RCIRNet outperformed other models; SVM showed stability, and KNN excelled in Recall.

Conclusions:

• Selecting discriminant movements is more critical for RCI recognition than increasing task quantity.
• The shoulder frontal ROM test combined with RCIRNet offers an optimal strategy for accurate and efficient RCI screening and monitoring.