Dorsal-to-Joint Temperature Differences (D2JTD) extraction from RGB-T images for remote rheumatoid arthritis monitoring

Rheumatoid arthritis (RA) is a chronic autoimmune disease characterized by progressive joint inflammation, presenting clinically with swelling, pain, and functional impairment. While clinical evaluation remains the diagnostic gold standard, we propose dorsal-to-joint temperature difference (D2JTD) as an adjunctive monitoring tool for disease activity, collected through the ”Living With RA” mobile application using smartphone-connected thermal cameras from 82 RA patients (stratified into low/moderate/high activity subgroups by RAPID3 scores) and 30 healthy individuals. Our methodology integrates multimodal RGB-thermal (RGB-T) image fusion with physical distance-assisted hand keypoint estimation (97.65% accuracy) and adaptive joint segmentation (Dice score 0.90±0.06) to calculate D2JTD values. Key findings demonstrate significant differences between RA patients (1.1^∘C±0.7) and normal group (1.5^∘C±0.4), particularly showing the lowest values in high-activity subgroups (0.8^∘C±0.3), along with reliable classification performance achieving 76.80% recall and 78.95% precision in distinguishing disease activity states. This temperature-based monitoring approach demonstrates potential for community screening and telemedicine applications, warranting further validation through multicenter clinical trials to explore its utility in treatment monitoring.