Deep learning model for classifying shoulder pain rehabilitation exercises using IMU sensor

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Artificial Intelligence in Healthcare

Background:

• Artificial intelligence (AI) is increasingly used in rehabilitation for monitoring exercise compliance via sensor technology.
• Previous AI classification of shoulder exercises using IMU sensors was limited to pain-free subjects.
• This study addresses the need to classify shoulder pain rehabilitation exercises in patients.

Purpose of the Study:

• To classify 11 types of shoulder rehabilitation exercises using an AI algorithm in patients experiencing shoulder pain.
• To demonstrate the feasibility of monitoring exercise compliance in a clinical population.
• To validate the accuracy of AI-driven exercise classification.

Main Methods:

• Collected data from 58 patients (37-82 years) with shoulder conditions like adhesive capsulitis and rotator cuff disease.
• Patients performed 11 types of shoulder pain rehabilitation exercises 10 times each, wearing an IMU sensor.
• Utilized Rectified Linear Unit (ReLU) and SoftMax activation functions within a deep neural network (DNN) model.

Main Results:

• A deep neural network (DNN) model, employing a multilayer perceptron algorithm, was trained on the acquired sensor data.
• The trained model achieved a high training accuracy of 0.975 and a test accuracy of 0.925 in classifying exercises.
• Demonstrated effective classification of 11 distinct shoulder pain rehabilitation exercises.

Conclusions:

• IMU sensor data, processed by AI, can accurately classify shoulder pain rehabilitation exercises, offering better patient feedback.
• The developed model can support remote patient monitoring systems for exercise performance.
• Deep learning holds significant potential for innovating healthcare service delivery in patient monitoring and rehabilitation.