Predicting Type 2 Diabetes Mellitus Using Fundus Eye Scans: A Deep Learning Approach

Type 2 Diabetes Mellitus is a persistent metabolic disorder with increasing global prevalence, requiring prompt identification to prevent serious consequences. Conventional diagnostic techniques depend on intrusive blood testing, which may fail to detect early onset. Recent breakthroughs in deep learning and medical imaging have facilitated non-invasive methods for illness prediction. This work investigates the application of colour fundus photography and deep learning methodologies to predict Type 2 Diabetes Mellitus by identifying retinal biomarkers. We present an optimised Convolutional Neural Network trained on an extensive dataset from the AIREADI Project, utilising hyperparameter tuning and regularisation methods to improve generalisation. The model attained an accuracy of 84.0%, indicating its efficacy in differentiating between diabetes and non-diabetic patients. The results underscore the future potential of AI-enhanced fundus imaging as an efficient, economical, and accessible screening method for the early identification of T2DM, especially in resource-constrained environments.