Markerless motion capture-based machine learning models for low back pain classification: a proof of concept investigation

Low back pain affects up to 70% of individuals globally, posing significant health and economic challenges. Traditional methods for classifying low back pain, including radiological imaging and functional tests, are limited by the substantial costs and limited utility via radiological imaging, as well as the inherent subjectivity of functional tests. To address these limitations, we propose a movement-based screening framework that integrates deep-learning-based “Ergo” 3D markerless motion capture with machine learning for objective LBP classification during an overhead squat test. Seven machine learning algorithms, including logistic regression, random forest, support vector machine, k-nearest neighbors, multilayer perceptron, gradient boosting, and the random forest gradient boosting ensemble, were evaluated. All the machine learning algorithms achieved accuracy above 0.82, and the random forest gradient boosting ensemble model achieved the best performance (accuracy = 0.96, AUC = 0.99). Model interpretability was enhanced using SHAP and permutation importance, linking data-driven features to known biomechanical characteristics of LBP. Our results support the feasibility of a machine learning motion capture approach as a cost-effective, non-invasive, and objective screening tool for low back pain classification.