A ResNet-based approach for accurate radiographic diagnosis of knee osteoarthritis

Abstract

Currently X-ray images are clinically graded by experienced clinicians using the Kellgren and Lawrence (KL) scoring method. However, individual scoring is subjective and error prone. This study proposes an approach for automated knee osteoarthritis classification based on deep neural networks. The knee X-ray images are first pre-processed with frequency-domain filtering and histogram normalisation, making the trabecular bone texture more obvious and benefiting the subsequent classification task. Then, a two-step classification strategy is proposed by extracting the joint centre based on the VGG network and classifying osteoarthritis grades based on the ResNet-50 network. In addition, a rebalance operation is proposed to deal with the dataset unbalance problem, and a quick search technique is proposed to improve the iterative search efficiency for the joint centre. With all of these techniques, a classification accuracy of 81.41% is obtained, which is higher compared to the state-of-the-art approaches.