Classification of knee osteoarthritis severity using markerless motion capture and long short-term memory fully convolutional network

Résumé fourni par la source

This study explored the integration of markerless motion capture and deep learning to classify knee osteoarthritis severity based on gait kinematics, providing an alternative to traditional assessment methods. We employed a Long Short-Term Memory Fully Convolutional Network model to analyze gait patterns and distinguish severity levels corresponding to Kellgren-Lawrence grades. Two data-splitting strategies were compared: random splitting, which allowed learning from a diverse set of individuals, and subject-based splitting, which assessed the model's ability to generalize to unseen patients. The model achieved high classification performance in the random split (accuracy = 0.91), but performance declined in the subject-based split (accuracy = 0.76), indicating challenges in generalizing across individuals. Severe and healthy groups were well classified, while early and moderate severity groups showed higher misclassification rates due to overlapping gait characteristics. Our findings highlight the potential of gait-based deep learning models for automated severity classification, offering a scalable and accessible alternative to conventional assessments. However, challenges related to inter-subject variability suggest the need for enhanced feature extraction, multimodal data integration, and domain adaptation to improve generalizability. Future research should focus on longitudinal data to assess the model's predictive capability for disease progression and treatment outcomes.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.