Early Detection of Diabetic Retinopathy Using Pretrained Models: A Focus on Initial Stages

Early detection of diabetic retinopathy is crucial because this disease currently has no cure. Identifying the condition at an early stage can significantly help in halting its progression and preventing severe vision loss. Numerous studies have focused on the detection and classification of diabetic retinopathy using various machine learning and deep learning techniques. In this work, we aim to detect the disease in its initial stages using a pre-trained VGG16 model, which is proved to be efficient in many image recognition tasks by recent researchers. Our initial approach achieved an accuracy of 79%, indicating a promising potential for early detection. To further improve the detection accuracy, we incorporated Ben Graham's preprocessing method, which enhances the finer details of the retinal images. This enhancement led to a 5% increase in accuracy, resulting in a final accuracy of 84.40%. These results underscore the effectiveness of advanced preprocessing techniques combined with deep learning models in the early detection of diabetic retinopathy.