Diabetic Retinopathy Detection: A Transfer Learning Based Approach for Accurate Diagnosis
Chahil Choudhary, Avinash Mathur, Ruchika Gupta
Abstract
Diabetes patients run the risk of getting diabetic retinopathy, a serious and occasionally blinding eye condition. One of its distinguishing features is harm done to the bloodstream capillaries within the retina, which is light-sensitive tissue located at the back of the eye. Diabetes-related high blood sugar levels can cause anomalies in the blood vessels that manifest as leakage, edoema, and the growth of brand-new, delicate blood vessels. Artificial intelligence systems such as Deep Neural Networks (DNNs) have shown promise in the detection and diagnosis of diabetic retinopathy. DNNs can search through retinal images for the disease's specific symptoms, enabling early diagnosis and therapy. In this study, a recommended method based on transfer learning and a deep neural network-based architecture was employed to categorize diabetic retinopathy. The recommended model was trained using a set of retinal images from diabetic patients that were annotated with varying degrees of diabetic retinopathy severity. After model testing, a loss value of 0.8253 and 73% accuracy were achieved. This displays the model's ability to accurately classify retinal images into different levels of diabetic retinopathy severity. The acquired accuracy demonstrates how effectively the proposed model and suggested transfer learning technique manage the complexity of categorizing diabetic retinopathy.