An Automated Diagnosis of Diabetic Macular Edema Grading Using Transfer Learning

Diabetes affected millions of individuals till now, up from 108 million in past years, and the ratio is increasing daily. Age-specific diabetes mortality rates increased exponentially. Blindness is frequently brought on by diabetes. With medication, routine screening problems, and treatment for any that arise, diabetes can be controlled and its effects postponed or averted. Therefore, for earlier detection of diabetes-based disease, Diabetic Macular Edema is diagnosed using transfer learning deep models. The transfer learning is used on the pre- trained EfficientNetBO and ResNet50 models. The CNN layer and hyperparameters are added to these models to combat overfitting. Moreover, the dataset is well pre-processed and augmented using specific parameters to avoid misclassification and better model training. The work in this paper does categorical classification instead of binary classification, which makes the patient clearer about the severity of the disease. The models are trained and evaluated on MESSIDOR and IDRID diabetic macular edema publicly available datasets. This work has been observed to be more effective than other published work on this disease.