Diabetic Retinopathy Diagnosis with Image Processing

This study focuses on developing deep learning-based models and improving their performance through deep learning techniques for the early diagnosis of “Diabetic Retinopathy”, a disease that develops due to diabetes and may cause vision loss in the eye in later stages. Deep learning models were developed using image processing methods on the limited access “Brazilian Multi-Label Ophthalmological Dataset”. Deploying AlexNet model, classification of diabetic retinopathy, macular edema, scar, nevus, drusens, increased cup disc, myopic fundus and age-related macular degeneration made based on fundus photographs. Models were trained with three different approaches for diabetic retinopathy and other eye diseases that may be associated with diabetic retinopathy. The developed method is aimed to facilitate decision-making by expert ophthalmologists in the diagnosis of diabetic retinopathy and other eye diseases and to prevent possible misdiagnosis. Using sensitivity-precision parameters as model success metrics in unbalanced test data sets has increased the accuracy in diagnosing diseased data. For the balanced test data set, the diagnosis of diabetic retinopathy was predicted with an accuracy of 0.748 and an f1 score of 0.726.