Identification of ocular disease from fundus images using CNN with transfer learning

Abstract

Eye diseases are one of the serious health problems affecting human life. Detecting and diagnosing them early is critical to prompt treatment and preventing vision loss. However, all studies in the field of eye disease classification using machine learning models are limited to the detection of single diseases, and the accuracy rate is still low in multi-class systems. In this study, we propose a multi-class classification model using four pre-trained CNNs (DenseNet121, ResNet50, EfficientNetB3 and VGG16). The model classified eye diseases into four categories: diabetic retinopathy, cataract, glaucoma, and normal. To improve the training process, another data augmentation technique is applied to increase the amount of data. The performance metrics of the system are calculated using the confusion matrix. DenseNet-121 shows excellent performance in retinal disease classification in 30 epochs of training, with training and test accuracy reaching 99.97% and 96.21% respectively. The implementation of this system should be considered as a very useful means to help ophthalmologists to rapid and precision detection of various eye diseases in the future.