Reliability-aware fusion of ocular thermal imaging and radiometric features for proof-of-concept classification of diabetic retinopathy vs. healthy controls

The study introduces ReliaFusion-Net, a new system for detecting diabetic retinopathy using thermal eye imaging and temperature data. The research included 558 thermal eye images (278 healthy and 280 with diabetic retinopathy) collected with ophthalmologist support. The system combines advanced image processing techniques to analyze the relationship between thermal images and physiological temperature data. The model achieved 93.18% accuracy, 93.26% F1-score, and 0.9785 AUC on test data. Results demonstrate that this method can effectively distinguish eyes with diabetic retinopathy from healthy eyes without invasive procedures. This technology has potential for use in telemedicine and may serve as an alternative to expensive traditional screening methods.