Deep Learning Based Decision Support System for Retinal Disease Classification: Diabetic Retinopathy and Macular Hole

In this study, a deep learning-based decision support system was developed to classify diabetic retinopathy (DR), macular hole (MH) and healthy samples using fundus images. A total of 1,397 fundus images selected from the open source Retinal Disease Classification dataset were used in the training and testing phases. ResNet50, InceptionV3 and Xception models were trained with different hyperparameter configurations and their performances were comparatively evaluated. As a result of the analysis, the ResNet50 model showed the highest success on the test set with an accuracy of 93.79%. However, the Xception model stood out with its consistent performance against different hyperparameter combinations and provided the most balanced results in terms of average accuracy. The results show that deep learning-based methods can be effectively used as a clinical decision support system for retinal disease diagnosis.