Detection of Low Back Physiotherapy Exercises With Inertial Sensors and Machine Learning: Algorithm Development and...

Area of Science:

• Biomedical Engineering
• Rehabilitation Technology
• Machine Learning in Healthcare

Background:

• Physiotherapy is essential for managing low back pain (LBP).
• Objective measurement of physiotherapy adherence is needed for effective LBP recovery.
• Current methods lack quantitative assessment of unsupervised exercise performance.

Purpose of the Study:

• To develop and evaluate a wearable inertial sensor system for objectively detecting unsupervised LBP exercise and posture performance.
• To assess the system's ability to capture multi-planar movements and sitting positions.
• To establish a quantitative measure for physiotherapy adherence in LBP management.

Main Methods:

• A quantitative classification design utilizing machine learning (ML) was employed.
• Eight inertial sensors collected data during 7 McKenzie exercises and 3 sitting postures in healthy adults.
• Engineered time-series features and a convolutional neural network (CNN) were used to train and evaluate classification models (Random Forest, XGBoost).

Main Results:

• The optimal system used 3 sensors (lower back, left thigh, right ankle) with acceleration, gyroscope, and magnetometer channels.
• XGBoost achieved high F1 scores for exercise (0.94) and posture (0.90) classification.
• A CNN model demonstrated comparable performance using fewer sensor channels.

Conclusions:

• A 3-sensor wearable solution (e.g., smart pants) can effectively identify exercises and postures for LBP treatment.
• This technology can enhance LBP rehabilitation by providing quantitative feedback and enabling remote monitoring.
• Objective performance tracking has the potential to improve patient outcomes and facilitate early diagnosis.