A mechatronic and artificial intelligence-driven framework for automated non-invasive knee abnormality screening using multimodal sensor data

Current knee-abnormality detection relies on costly Magnetic Resonance Imaging (MRI) and subjective clinical evaluation, limiting accessibility. This study presents an integrated mechatronic and machine-learning framework using surface electromyography (sEMG) and goniometers to capture multimodal mobility data. Novel time-frequency features, Enhanced Mean Absolute Value (EMAV) and Enhanced Wavelength (EWL), improved Extra Trees accuracy by 3.16% over conventional MAV and WL. The Extra Trees classifier achieved cross-validated accuracy of 94.7%, with 95% precision and recall, validated by Friedman and Nemenyi tests. Shapley Additive exPlanations (SHAP) analysis adds interpretability. The framework provides a low-cost, automated screening solution extendable to human-movement and pathology applications.