Prediction of Diabetic Eye Disease in Type 2 Diabetes Using Hybridization of Deep Neural Network-Based Approaches

Diabetic Eye Disease (DED) notably affects the retina of human eyes. It is an ophthalmic complication generally found in diabetic patients that may cause blindness incase not diagnosed at an early stage. For experimentation, the researchers use large-sized datasets consisting of retinal images. The careful training of these retinal images is very significant for training deep learning models so that signs of diabetic eye disease can be identified accurately. The deep learning architectures used in this study, include, EfficientNetB7, DenseNet201, InceptionResNetV2, Xception, and ResNet101V2. These architectures can extract meaningful features from retinal images, that are required for classification tasks accurately. In addition to individual models, a combination of ResNet101V2 and DenseNet201 is considered to inherit the strengths of both models so that predictive performance can be enhanced. To assess the performance of each of the model, various quality measures, namely precision, F1 score, and recall are used. The obtained results clearly indicate the potential of the aforementioned hybrid model for predicting the diabetic eye disease in the human eye.