Abstract

Rheumatoid arthritis (RA) encompasses damage to joint tissues, and its pathology can be diagnosed using available medical imaging modalities like X-rays, MRI, and Ultrasound. It is one of the lifelong progressive autoimmune diseases that require early detection and prompt treatment to improve patient outcomes. This study proposes an integrated convolutional neural network-recurrent neural network (CNN-RNN) framework to facilitate RA diagnosis by jointly using spatial and temporal features from sequentially acquired medical images. The model was trained and evaluated by using over 3000 radiographic images acquired from 500 patients, including patient-wise split into training (70%), validation (15%), and testing (15%) subsets. Intensive preprocessing included contrast enhancement, noise reduction, and normalization of image quality. The CNN module extracts discriminative spatial features from individual frames, while the RNN component captures the progress of the disease across several time points. Experimental assessment showed that this integrated model achieves a classification accuracy of 94.8%, precision-93.6%, recall-95.1%, and F1-score of 94.3%, significantly superior to traditional CNN- and SVM-based methods. The statistical significance was established using McNemar's test (p<0.01), thus confirming the superiority of the proposed approach. Together, these findings emphasize the potential of deep spatial-temporal learning towards the more accurate and earlier detection of RA in clinical settings.