Synergizing Generative Adversarial Networks and Pseudo-Labeling for Improved Knee Osteoarthritis Detection

Progressive cartilage breakdown and joint inflammation are hallmarks of the common joint disease known as osteoarthritis (OA). The timely intervention and management of osteoarthritis (OA) depend on early detection. However, this can be difficult to achieve because labelled medical imaging data is scarce, and the disease's presentations are complex. A notable skew between the proportion of healthy and OA knee images can be seen in knee OA datasets, a phenomenon known as class imbalance. As a result, the model may underperform on the minority class (OA) and prioritize the majority class during training. The goal of the research is to further the development of more accurate and reliable techniques for knee OA detection. A novel method for knee osteoarthritis detection that first uses ResNet, DenseNet, VGG16, and VGG19 convolutional neural network (CNN) architectures for classification after generative adversarial networks (GANs) for data augmentation and pseudo-labeling is predicted. The aim of the work is to use labeled and unlabeled data to develop the robustness and accuracy of osteoarthritis detection. It is shown via thorough experimentation that the strategy is beneficial in enhancing classification performance, with ResNet obtaining the greatest accuracy and F1 score among the networks we studied. According to the research, pseudo-labeling and GAN-based data augmentation strategies can greatly improve osteoarthritis diagnosis accuracy and clinical significance. This work advances the field of medical image analysis and has potential benefits for bettering osteoarthritis patient treatment. The proposed approach performs well with an accuracy of 96.23%, precision of 0.963, recall of 0.959, and F1 score of 0.9246.