Explainable Optimizing Knee Osteoarthritis Diagnosis: Leveraging Multi-label Learning and Clinical Features for Enhanced Accuracy

Knee Osteoarthritis (KOA) is a widespread degenerative joint condition impacting millions worldwide, resulting in gradual decline over time and hindering patients’ daily functions due to persistent pain and discomfort. Despite recent strides in Artificial Intelligence facilitating automated diagnosis using X-ray or MRI scans, many current approaches neglect the role of clinical variables in predictions, relying solely on imaging data for classification. In this study, we introduce an innovative methodology for evaluating KOA severity that incorporates both medical images and clinical information. We focus on applying multi-label classification to medical images aiming to generate a more comprehensive prediction compared to solely assessing disease severity. By incorporating multi-label classification, we can identify the presence of additional vital risk factors. To further enhance the prediction’s accuracy, we also integrate relevant clinical information with the image analysis. We then showcase the efficacy of our approach by outperforming existing methods. Furthermore, our model sheds light on the patient risk factors like joint space narrowing, osteophytes, and sclerosis through multi-label classification, rendering it using an explainable AI technique. To be more precise, each prediction from our model is accompanied by an explanation, thereby augmenting the interpretability of the diagnostic process.