Diabetic Retinopathy Detection and Severity Grading Using Deep Learning

Diabetic Retinopathy (DR) is a cause of blindness that necessitates timely detection [1], [2]. Manual diagnosis has the drawbacks of being time and effort-intensive with poor reproducibility, resulting in the quest for automated methodologies. In this work, the InceptionV3 model has been used based on deep learning, fine-tuned for classification of DR severity. The dataset used for training is APTOS 2019, in which preprocessing operations such as CLAHE, normalization, and data augmentation were employed to enhance detection accuracy [5], [6]. For improved accessibility, we created a web application based on Streamlet, enabling real-time DR severity prediction from uploaded fundus images. The system has an accuracy of 76%, showing its effectiveness as a diagnostic tool [11,18]. Some future improvements include lesion segmentation, detection of multiple diseases, and mobile deployment for further usability, especially in low-resource settings [22], [23].