Computer-Aided System for the Detection of Rheumatoid Arthritis

Rheumatoid Arthritis (RA) is a chronic disease that causes disability in movement. RA classification is critical for effective diagnosis and treatment planning. This work explores the application of the EfficientNetB6 architecture using transfer learning toclassify RA severity into three categories: Healthy, Moderate,and Severe. Medical imaging dataset containing X-Ray images, enhanced with contrast-limited adaptive histogram equalization (CLAHE), data augmentation techniques,and fine-tuning of hyperparameters wasappliedinthiswork.WecomparedEfficientNetB6withallthemodelsofthe EfficientNet family and all other state-of-the-art models. When we combined EfficientNetB6 with CLAHE, we achieved the highest accuracy of 96.06%. Without using CLAHE,accuracy dropped by 4% to 5% for all the models. For a healthy class model, we achieved precision, recall,and F1-score of 99.36%,97.81%,98.58% respectively, showing robustness in identifying healthy cases. Moderate class yielded precision, recall,and F1-score of 89.45%,95.07%,92.17%respectively,demonstratingthemodel’seffectivenessinidentifying moderate cases with minimal false negatives. The Severe class presented more challenges with a precision, recall,and F1-score of 85.11%,78.43%, 81.63% highlighting the need for improved recall value. To further improve results, we suggest enhancements such as advanced data augmentation and synthetic data generation, particularly for the Severe class,consequently aiding clinicians inthe identification of RA.