Classification of the Pathological Range of Motion in Low Back Pain Using Wearable Sensors and Machine Learning
Fernando Villalba-Meneses1,2,3, Cesar Guevara4, Alejandro B Lojan2
1IDERGO (Research and Development in Ergonomics), I3A (Instituto de Investigación en Ingeniería de Aragón), University of Zaragoza, C/Mariano Esquillor s/n, 50018 Zaragoza, Spain.
Sensors (Basel, Switzerland)
|February 10, 2024
Summary
This study developed a medical test using machine learning and motion capture data to aid in physical treatment decisions for low back pain (LBP). Support Vector Machine models achieved over 90% accuracy, offering a data-driven approach to improve patient quality of life.
Area of Science:
- Biomedical Engineering
- Machine Learning in Healthcare
- Musculoskeletal Disorders
Background:
- Low back pain (LBP) is a prevalent musculoskeletal condition causing significant work absenteeism.
- Current diagnostic methods for nonspecific LBP lack precision in guiding physical treatment.
- Motion capture (MoCap) technology offers objective biomechanical data for clinical assessment.
Purpose of the Study:
- To develop a machine learning-based medical test for guiding physical treatment in patients with nonspecific LBP.
- To classify patients with LBP using motion capture data from range of motion (ROM) exercises.
- To enhance the clinical applicability of MoCap for data-driven LBP management.
Main Methods:
- Collected motion capture (MoCap) data during ROM exercises from healthy individuals and patients with LBP in Imbabura, Ecuador.
- Trained and evaluated seven machine learning (ML) algorithms: logistic regression, decision tree, random forest, SVM, KNN, MLP, and gradient boosting.
- Compared algorithm performance based on classification accuracy for LBP detection.
Main Results:
- All tested ML algorithms achieved >80% accuracy in classifying LBP.
- Support Vector Machine (SVM), random forest, and MLP models demonstrated superior performance with >90% accuracy.
- The SVM algorithm was identified as the most effective for this classification task.
Conclusions:
- Machine learning models effectively classify LBP using MoCap data, achieving high diagnostic accuracy.
- SVM, random forest, and MLP show significant potential for developing objective LBP assessment tools.
- This data-driven approach using MoCap can improve the quality of life for individuals with chronic LBP.
