Optoelectronic Retinal Images for the Prediction of Diabetic Macular Edema Based on a Hybrid Deep Transfer Learning Technique

Diabetes is characterized by elevated levels of glucose in the blood, which can lead to complications like Diabetic Macular Edema (DME), causing permanent vision loss.A novel HET-EYE-NETS which is built on the ensemble transfer learning networks with extreme feedforward model for the prediction of DME is proposed.The proposed algorithm preprocesses the color optoelectronic retinal images and classifies the severity of DME by the three-stage pipeline model.In the first stage, DME is segmented by the U-Nets, features of segmented DME are extracted by AlexNet layers and finally severity is predicted by the extreme learning feedforward layers.The extensive experimentation is carried out using IDRiD and MESSIDOR database images.During this process, performance measures like precision, recall, F1-score, specificity, and accuracy are computed and analyzed.In addition, data augmentation is employed to address the issue of data imbalance problem in IDRiD and MESSIDOR database images.The proposed HET-EYE-NETS model achieved an average accuracy of 99.1%, precision of 99.2%, recall of 99% and F1-score of 0.9920.Results proved that proposed HET-EYE-NETS model outperforms existing learning models, demonstrating its potential for early diagnosis of DME.