Detection Of Diabetic Retinopathy Using Convolutional Neural Network

Abstract

Diabetic Retinopathy (DR) Is A Leading Cause Of Vision Loss Globally, Particularly Among Individuals With Long-standing Diabetes. Early Detection And Grading Of DR Are Vital To Prevent Irreversible Blindness. This Project Presents An End-to-end, AI-powered Web Application For The Automatic Classification Of Diabetic Retinopathy From Fundus Images Using A Convolutional Neural Network (CNN) Deployed Via A FastAPI Framework. The Trained Model, Based On TensorFlow, Classifies Input Retinal Images Into Five Categories: No DR, Mild, Moderate, Severe, And Proliferative DR. The Dataset Used To Train The Model Was Sourced From Kaggle, Consisting Of High-resolution Retina Images Labeled By Clinical Experts. The Application Provides A Modern, Responsive Frontend Using HTML, CSS, And JavaScript, Allowing Users To Upload Retinal Images And Receive Real-time Diagnostic Predictions. The Backend Model Preprocesses Uploaded Images Using OpenCV And NumPy, Resizing Them To 224x224 Pixels, Normalizing Them, And Feeding Them Into The Trained CNN Model. The System Aims To Serve As A Fast, Reliable, And Accessible Tool To Assist Ophthalmologists And Healthcare Professionals In Screening For DR. It Can Also Act As A Valuable Aid In Regions With Limited Access To Medical Infrastructure, Where Regular Eye Checkups Are Not Always Feasible. Through This Project, We Demonstrate The Real-world Application Of AI And Web Development For Medical Diagnostics, Bridging The Gap Between Complex Deep Learning Models And User-friendly Interfaces.

Keywords