Radiographic imaging-based joint degradation detection using deep learning

Osteoarthritis (OA) is a degenerative joint disease that primarily affects the knee joint. Currently, OA diagnosis relies on the examination of plain radiographs, a method susceptible to subjectivity and time consumption. This study aims to automatically assess the severity of knee OA using the "Kellgren and Lawrence (KL) grading system" based on plain X-rays. Leveraging 1650 digitized knee X-ray images, we implemented a custom MobileNetV2 architecture for a convolutional neural network with four distinct orientations. The methodology comprises two models: a fixed base and a trainable head. The MobileNetV2 network serves as the base model, while the proposed head model architecture includes an average pooling layer followed by a fully connected layer to enhance network efficiency. Results indicate that, except for grades 1 and 2, the methodology correctly identified KL grades with an accuracy of over 90%. Overall, the proposed approach demonstrates promising potential for classifying knee OA based on plain X-rays, achieving a 95% accuracy in detecting severe knee OA (stage 4). Researchers acknowledge the superior performance of their methodology compared to previous models in similar investigations, suggesting its effectiveness in forecasting OA severity based on radiographic imaging. Furthermore, the study's results support the effectiveness of deep learning-based approaches in diagnosing OA severity, with significant implications for improving patient outcomes. Although the suggested methodology shows a maximum accuracy rate of 95% in identifying severe knee OA cases (specifically, stage 4), there is a need for enhancements to address the misdiagnosis issue in stage 1 and 2 knee OA, where the accuracy rate is at 87%. This misdiagnosis arises from the similarity in characteristics between these stages.