A Classification Framework for Diabetic Retinopathy Detection Using Transfer Learning

Diabetic Retinopathy is a serious eye condition resulting from long-term diabetes mellitus that can lead to permanent blindness if not treated on time. Early detection can reduce the likelihood of serious disability. This research introduces a binary classification framework using Transfer Learning to identify diabetic retinopathy in diabetic patients. The research makes use of an image-based dataset, APTOS 2019, sourced from Kaggle. These images are used to train transfer learning models for predicting the presence or absence of this disease among patients. Preprocessing steps including resizing, Gaussian filtering and dataset split are employed prior to classification. Afterwards, data augmentation techniques like rotation, zooming, flipping, shifting and rescaling are applied to increase the training set. For classification, two Transfer Learning models, EfficientNetB3 and VGG16, are used after fine-tuning. The results indicate that EfficientNetB3 model outperformed VGG16 model due to its computationally efficient architecture, achieving a test accuracy of 97.82%.