A Hybrid Model for the Prediction of Diabetic Retinopathy in Type 2 Diabetes

Diabetic Retinopathy(DR) is an ophthalmic ailment that is generally found in Type 2 Diabetic patients. This disease causes notable effect on the retina of the human eyes. The classification of neural network models and diagnosis of DR at an early stage can avert the progression of disease. The neural networks models need to be trained for the classification as well as detection task. However, a deep neural network requires a significant volume of training data and sufficient time to accomplish the training of the model. Therefore, in this research work a hybrid model has been proposed for the prediction of diabetic retinopathy to resolve aforementioned issues, while retaining high accuracy. In the initial stage, standard datasets for diabetic retinopathy and blindness are obtained. Then, DenseNet201, ResNet101V2, InceptionResNetV2, EfficientNetB7, and Xception CNNs are chosen as the base models. The proposed hybrid model structure (DenseNet201 + ResNet101V2) is trained using the output obtained from the base model. In the final stage, the results of the hybrid model are compared with existing base models. The performance of proposed hybrid model is evaluated on the basis of various quality measures, such as accuracy, F1 score, recall, precision, loss, RMSE values, and loss. According to the experimental results, the proposed hybrid model (DenseNet201+ResNet101V2) achieved 98.37% classification accuracy of on the training datasets whereas 93.25% on the validation datasets.