Rheumatoid Arthritis Predictor Using ML Techniques and Explainable AI

Abstract

Rheumatoid Arthritis (RA) is a chronic autoimmune disease that occurs in multiple organs and joints in the body. It is characterized by inflammation in the lining of joints, known as the synovium, which leads to pain, stiffness, and eventually, loss of function. RA can also cause fatigue, fever, and weight loss, and in severe cases, it can lead to permanent joint damage and disability. Diagnosis is essential during the early stages of affection, but the current process is costly and inefficient, disadvantaging those with limited finances. To address this, a study was conducted using five different machine learning (ML)models (Convolutional Neural Networks (CNN), K-Nearest Neighbor (KNN), Xg-boost (XB), Gaussian Naive Bayes (GNB), and Support Vector Machine (SVM)) to find the most efficient way to diagnose RA. The study compares the accuracy of these algorithms and determines the most effective one for predicting RA in a patient. The process involves two main steps: image processing and algorithm-based prediction. During the image processing phase, the uploaded image undergoes optimization techniques to remove false negatives and enhance the image quality for a more ideal input to the following step. The image is processed using the most effective ML model in the second step, which results in 98% prediction accuracy, a significant improvement over and above the state-of-the-art literature.