Augmenting Diabetic Eye Disease Detection Using Advanced Deep Neural Network

Diabetic eye diseases inclusive of Diabetic Retinopathy (DR), Diabetic Macular Edema (DME), glaucoma, and cataract, constitute significant contributors to the global pool of vision impairments. This is why for preventing vision lose, early detection plays a crucial role as well as efficient detection. This site presents a very elaborate deep neural network (DNN) framework is presented, wherein the integration of Vision Transformer (ViT) and EfficientNetV2 acts to increase the detection and classification of the above thought diseases. The model employs EfficientNetV2 for the extraction of local spatial features, while ViT acts to capture global contextual information. To ensure that representations of the features are considerably enhanced, cross-attention fusion is used to fuse outputs. To counterclass imbalance, we used Focal Loss and advanced augmentation methods such as MixUp and CutMix six public data sets are used to test the performance of the model fundus datasets (DIARETDB0, DIARETDB1, Messidor, HEI-MED, Ocular, Retina) and measured using as performance indices, accuracy, precision, recall, AUC, and F1-score have been used. From the results of the experiments, it can be inferred that this model is outperforming any existing modelsbased on CNNs, as it exhibits excellent classification and reliable detection results in all four classes of disease. This hybrid framework shall be seen as a big step toward the automated diagnosis of diabetic eye diseases with a reliable and scalable solution for timely detection of diseases and enhancing the health of patients.