Rheumatoid arthritis (RA) automated diagnosis is essential for early identification and successful therapy planning. A deep learning-based RA diagnostic paradigm that combines medical imaging, blood biomarker analysis, and sophisticated image processing algorithms is presented in this paper. In the suggested method, picture patches are extracted using a Supervised Adaptive Noise Reduction Convolutional Neural Network (SANR_CNN) for denoising, and a dictionary building process is used to improve image quality. In order to precisely identify joint structures, a Discrete-MultiResUNet model is then used for segmentation, which includes boundary selection, border correction, bone tissue segmentation, and masking. To ensure accurate RA severity grading, the last step entails classifying RA severity using a ResNet50 model improved using the Adam optimizer. The performance of feature extraction, noise reduction, and classification is improved by combining various deep learning approaches. The suggested approach is a potential tool for automated RA detection and classification in clinical applications, as evidenced by experimental validation showing that it performs better in terms of accuracy, precision, and specificity than conventional RA diagnostic techniques.