Diagnostic Value of Knee Osteoarthritis Through Self-learning

Osteoarthritis (OA) is the most common chronic and progressive musculoskeletal disorder. These chronic disorders are not diagnosed early nor is the treatment adequate, resulting in challenges for health care systems. Radiography is the most widely used imaging method for OA diagnosis since it allows a two-dimensional evaluation, whose main advantages its low cost and wide availability. We present a computer-assisted diagnostic (CAD) system for OA based on the analysis of X-ray images of the knee using deep learning to automatically score the Knee OA. The scoring is based on the Kellgren-Lawrance (KL) scale. The model was implemented in PyTorch and was based on Deep Siamese convolutional neural networks and fine-tuned ResNet-34 through transfer learning for the classification task. A public dataset was used for training and validating, and a private dataset for testing. The results indicate a multiple-class accuracy of the test set of 61%. The highest accuracy was obtained with KL-3 at 89%. It is expected that this software will be useful for training of medical students and can be used as a second opinion for the correct prediction of OA knee diagnosis. Early diagnosis is necessary to alleviate symptoms, delay the evolution of the disease and improve the functional capacity and quality of life of the patient.