A Tool for Low-Cost, Quantitative Assessment of Shoulder Function Using Machine Learning.

Area of Science:

• Biomechanics
• Medical Diagnostics
• Computational Biology

Background:

• Rotator cuff (RC) tears are a leading cause of shoulder pain, particularly in older adults.
• Current diagnostic methods like imaging are expensive and require in-person examination.
• There is a need for accessible, low-cost methods to assess shoulder function and diagnose RC tears.

Purpose of the Study:

• To investigate a simple string pulling task as a reliable indicator of shoulder health.
• To identify quantitative biomarkers of rotator cuff tears from movement kinematics.
• To develop a predictive model for diagnosing rotator cuff tears using movement data.

Main Methods:

• Subjects performed a hand-over-hand string pulling task.
• Movement kinematics (amplitude, time, waveform shape) were analyzed in mice and humans.
• A machine learning model was trained on movement biomarkers to classify RC tear status.
• The model's accuracy in classifying human RC tears was evaluated.

Main Results:

• String pulling task performance revealed decreased movement amplitude and prolonged movement time in individuals with RC tears.
• Quantitative changes in movement waveform shape were observed in both animal models and human patients.
• A predictive model achieved >90% accuracy in classifying human patients with RC tears.
• Degradation of low-dimensional, temporally coordinated movements was noted in rodent models.

Conclusions:

• A string pulling behavior task provides a reliable readout of shoulder health and detects rotator cuff tears.
• Movement kinematics combined with machine learning can accurately diagnose shoulder injuries.
• This framework enables the development of smartphone-based, at-home diagnostic tests for shoulder injuries.