Retinal fundus image classification for diabetic retinopathy using transfer learning technique

Diabetic Retinopathy (DR) stands as a primary cause of blindness across all age groups, attributed to insufficient blood supply to the retina, retinal vascular exudation, and intraocular hemorrhage. Despite recent strides in DR diagnosis and treatment, this complication remains a formidable challenge for both physicians and patients alike. Consequently, the demand for a comprehensive and automated DR screening approach has become imperative, aiming to achieve early detection and potentially revolutionize the management of this disease. This study introduces a novel approach for identifying diabetic retinopathy through transfer learning-based optical image data classification. We have proposed four methods based on pretrained models: VGG16, VGG19, InceptionV3, and DenseNet169. The effectiveness of the newly reformed networks is evaluated using four performance metrics, using the APTOS2019 dataset as the basis for validation. The results demonstrated that the InceptionV3 model achieved the highest accuracy of 96.88%. It outperformed all other state-of-the-art diabetic retinopathy detection models.