Challenges and Solutions with Lightweight Models for Diabetic Retinopathy Detection

Diabetes retinopathy (DR) will become the primary cause of blindness in developing nations if it is not properly managed at the appropriate time. A cost-effective method for screening a large number of people with diabetes with DR identification worldwide appears to be a challenging task. Achieving the mass screening of DR in an efficient and effective way can be made possible by applying suitable deep learning algorithm models with impressive results of high sensitivity and specificity in fundus photographs that can run on handheld devices. The research work focuses on addressing the challenges in diabetic retinopathy (DR) detection by conducting experiments on various datasets. The novelty of the proposed system lies in the use of simple CNN and transfer learning VGG16 models to achieve accurate results with different datasets. The study also explores the merging of datasets to create a lightweight model suitable for implementation on handheld devices. The research highlights the importance of refining and tuning the models to improve accuracy and reduce model size, ultimately benefiting mass screening for DR prediction. Medical advances would then result from the development of AI technology, and doctors and ophthalmologists if carefully used these technical advancements will definitely benefit society.