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

Comprehensive Summary

This study integrated markerless motion capture with a deep learning (DL) model to classify the severity of knee osteoarthritis (OA) based on gait kinematics. A Microsoft Kinect system was used to collect gait data, which was then processed to reduce noise while maintaining key features. A Long Short-Term Memory Fully Convolutional Network (LSTM-FCN) analyzed the gait patterns of participants and classified OA severity based on the Kellgren-Lawrence scale. The model achieved an accuracy of 0.91 under random splitting and 0.76 under subject-based splitting, with misclassifications most frequent in early and moderate severity groups.

Outcomes and Implications

Knee OA is a leading cause of pain and reduced quality of life in older adults. It often results in impaired joint function and altered gait kinematics. Markerless motion capture paired with a DL model is a time-efficient, low-cost alternative to previous gait kinematic analysis systems. This model could lead to improved diagnosis accuracy and promote earlier intervention in patients with knee OA. However, generalizability is still a concern for the model. Improvement of feature extraction and external validation is needed before clinical integration.