Detection of Diabetic Retinopathy by Segmentation using U-Net with Hyper-Parameter Tuning

Abstract

Diabetic Retinopathy (DR) is a retinal disease that occurs in Diabetes Mellitus (DM) patients due to damage to the retinal blood vessels. Among the micro and macrovascular, DR is the major complaint. The effect of DR can range from impairment of vision to loss of vision based on its severity. Ophthalmic complications of uncontrolled DM are dry eye, recurrent infection of the lid, Diabetic macular edema, cataracts, and glaucoma. Diagnosing the disease by detecting the lesion is to be done to find the severity level or grades of DR. Various automatic approaches based on deep learning are available to detect the lesions. In this paper, the U-Net algorithm with hyperparameter tuning was used to identify the lesions present in the retina and obtained an accuracy of 91.23%, 92% of sensitivity, and 93.37% specificity with the private dataset