ML Application

Diabetic Retinopathy Classifier

End-to-end medical AI with web deployment on AWS

Fine-tuned VGG16 to classify 5 severity grades of diabetic retinopathy from fundus images. Built a Flask web interface and deployed on AWS EC2 for clinical accessibility.

View on GitHub

Tech stack

TensorFlowKerasVGG16FlaskOpenCVAWS EC2

Problem

Diabetic retinopathy is a leading cause of blindness, but grading severity requires specialist expertise. An automated classifier can triage cases and flag urgent referrals.

Implementation

▸VGG16 (ImageNet pre-trained) fine-tuned on Kaggle Diabetic Retinopathy dataset

▸5-class classification: No DR, Mild, Moderate, Severe, Proliferative DR

▸Data preprocessing pipeline: CLAHE enhancement + augmentation (flips, rotations, colour jitter)

▸Flask web app with image upload and confidence score display

▸Deployed on AWS EC2 with Gunicorn + Nginx

Stack

TensorFlow/Keras · Flask · OpenCV · AWS EC2 · HTML/CSS

All projects