Enhance Diabetes Management using AI and Machine Learning for Glycemic Control

Diabetic Retinopathy is a serious, progressive eye disorder that results due to damage to the retinal vasculature. It usually occurs in diabetic people due to long-term high blood sugar levels. Therefore, much emphasis has been done on developing and assessing sophisticated deep learning methodologies that can use retinal images for automatic detection of diabetic retinopathy. A dataset curated from Kaggle is used as the core, which contains 1,920 MRI-based retinal scans of different stages of DR. The dataset consists of five classes: Normal retina (Stage 1), Proliferative diabetic retinopathy (Stage 5), mild non-proliferative (Stage 2), moderate non-proliferative (Stage 3), and severe non-proliferative (Stage 4). This allows for structured classification and deep-level analysis. Further, EnlightenGAN is used to enhance the sharpness and visual detail to enhance the clarity of the images. Besides this, several highperformance deep learning architectures are used for image classification, namely CoaTNet and EfficientNet-B7. In addition, BRISQUE will be used without any more extra reference images to assess the effectiveness of augmented image quality. The project will allow the field of computer-aided ophthalmic diagnostics to get an efficient and scaled-up solution for early detection and management of diabetic eye diseases with the help of modern advancements in deep learning.