Intelligent Hybrid Deep Learning System for Automated Retinal Disease Diagnosis

Diabetic Retinopathy (DR) is a main cause of acquired blindness and a leading cause of preventable blindness in the world. Individual detection of DR on the embodiment of a retina is critical to timely intervention and save sight. This proposed project will utilize a Hybrid Deep Learning Model (resnet50 plus vision transformer (ViT)) with the aim of discriminating Diabetic Retinopathy (DR) and Non-Diabetic (Normal) cases. The ResNet50 element is efficient in obtaining local retinal specific characteristics including microaneurysms and hemorrhages, whereas the ViT element assesses overall structural details of the retina which allows full image insights. The combination of these two feature extractors promotes the discriminatory capability and strength of the model. The proposed model performed at 99.49% overall classification accuracy using the given dataset of more than 500 well-crafted images per class with the best precision, recall, and F1-score. The system is convenient to the clinical screening and tele-ophthalmology uses because of a Flask-based web interface that permits real-time image upload and instant disease detection.