Efficient Eye Disease Detection: A Deep Learning Approach with InceptionResNetV2 for Retinal Image Classification

The most serious reasons for human vision impairment and blindness include cataract, glaucoma, and diabetic retinopathy. All of them are believed to require early diagnosis so as not to cause irreparable damage and to be able to provide the necessary medicinal intervention on time. This paper presents an automated classification system on a retinal image with the InceptionResNetV2 deep learning model for the detection of such diseases. In the proposed model, a total of 4,177 images from the four normal, cataract, glaucoma, and diabetic retinopathy classes were used in a balanced manner for training and testing. Various augmentation techniques were applied to this dataset in order to enhance generalization and avoid the problem of overfitting. The overall accuracy of the model is 92.18%, where diabetic retinopathy is correctly classified with precision/recall equal to 1, and cataracts/glaucoma are also classified very precisely. These have further pointed out the challenges in differentiating the cases of glaucoma from normal ones and show areas where refinement is required. This study has demonstrated the potential use of deep learning in automation for the detection of eye disease and forms a basis for future clinical manifestations.