Detection and Grading of Diabetic Retinopathy in Fundus Images Using Transfer Learning Models

Abstract

Diabetic Retinopathy (DR) arises from long-term Diabetes Mellitus, leading to newline damage in the retina. DR condition is categorized into two types: Non-Proliferative DR newline (NPDR) with Normal, Mild, Moderate and Severe stages and Proliferative DR (PDR) is an newline advanced stage with abnormal retinal blood vessels. Although DR may initially show no newline symptoms, it can progress to severe visual impairment over time. Eyes affected by DR newline exhibit features such as microaneurysms, hemorrhages, exudates and abnormal blood vessel newline growth. Early detection and preventive measures for DR can help control or prevent further newline damage to the retina. This research aims to create an efficient DR diagnosis system using newline advanced technologies like Deep learning, Transfer Learning, Transformer learning and newline Optimization. newline The proposed framework comprises four main objectives: The initial objective newline is to identify and detect red lesions to determine the mild and moderate stages of DR in newline fundus images. The initial phase involves pre-processing to enhance the image quality, newline followed by blood vessel segmentation using Deep Dense_UNet model. The lesion newline segmented images generated from the red lesion detection were utilized as an input for newline training and categorization using the SE-ResCA-GTNet model. This approach achieved newline superior accuracy in distinguishing between grade 1 (mild) and grade 2 (moderate) stages newline