Integrated Xception- Random Forest Model for the Detection of Rheumatoid Arthritis in Hand Thermograms

Rheumatoid Arthritis (RA) is a persistent inflammatory condition that affects the finger joints, wrist, knee, and foot. It leads to severe disease progression and immobility if not treated in the earlier stage. The study aims to build a hybrid convolutional neural network (CNN) - machine learning (ML) model for classifying RA and healthy participants. The main objectives of the current study are to (i) develop an automated feature extraction method by using modified Xception features (ii) build an integrated model by combining Xception- Random Forest classifier for the prediction of RA disease. The database for the proposed work was created by acquiring thermal hand images of 50 healthy and 50 RA patients. A modified Xception model was employed to extract the features; feature selection was performed using a correlation heat map and classification using a random forest (RF) classifier. The RF classifier obtained an accuracy of 95.83 %, a sensitivity of 91.67 %, and a specificity of 100 %, respectively. Therefore, thermal imaging can be used as a supporting modality that can help the physicians to diagnose RA at the earlier stages.