Early-Stage Diabetic Retinopathy Prediction

Proliferative Diabetic Retinopathy - The system detects Proliferative Diabetic Retinopathy (PDR), the most advanced stage of the disease.
Mild Diabetic Retinopathy - an early stage of the disease where small microaneurysms begin to appear in the retina
No Diabetic Retinopathy - indicating a healthy retina with no visible signs of damage.

Early-Stage Diabetic Retinopathy Prediction Using Advanced Machine Learning

Project Overview Diabetic Retinopathy (DR) is one of the leading causes of preventable blindness among diabetic patients worldwide. Early detection of Diabetic Retinopathy is essential to prevent vision loss and reduce treatment costs.

This project develops an AI-powered system that analyzes retinal fundus images to detect early-stage Diabetic Retinopathy using advanced deep learning techniques. The system applies multiple convolutional neural network (CNN) models and handles class imbalance commonly present in medical datasets.

The solution includes a machine learning pipeline, a backend API for inference, and a frontend interface to upload retinal images and obtain predictions.

Objectives Detect early-stage Diabetic Retinopathy using retinal images. Apply advanced deep learning models for medical image classification. Handle class imbalance in medical datasets. Provide an AI-powered decision support system for DR screening. Develop a deployable system using FastAPI backend and web frontend.

Dataset This project uses the APTOS 2019 Blindness Detection Dataset. Dataset Source: https://www.kaggle.com/competitions/aptos2019-blindness-detection/data The dataset contains retinal fundus images labeled according to the severity of diabetic retinopathy.

Class Labels Label Description 0 No Diabetic Retinopathy 1 Mild DR (Early Stage) 2 Moderate DR 3 Severe DR 4 Proliferative DR Images are high-resolution retinal scans used for medical diagnosis.

Machine Learning Pipeline

Image Preprocessing

Resize images to 224×224 Data augmentation Normalization Deep Learning Models

The system uses a multi-model architecture: ResNet50 Efficientnet-B0

Built With

efficientnet-b0
fastapi
html/css
kaggle
opencv
python
resnet50
tensorflow/keras

Submitted to

Code With Gemini

Created by

independently developed the Early-Stage Diabetic Retinopathy Detection System as an end-to-end AI application.
Dataset Handling: Collected and worked with the APTOS 2019 dataset, performed data cleaning, preprocessing, image resizing, normalization, and handled class imbalance to improve model performance.
Model Development: Designed and trained deep learning models using CNN architectures such as ResNet50 and EfficientNet for accurate classification of retinal images.
System Development: Built the complete pipeline including model training, evaluation, and integration into a backend using FastAPI.
Frontend Development: Developed a user-friendly web interface to upload retinal images and display real-time predictions with confidence scores.
Integration: Connected the frontend, backend, and machine learning model to create a fully functional application.
Documentation: Prepared the project report, including methodology, results, and analysis.
Overall, I handled the entire project lifecycle, from data collection and model building to application development and final deployment.

Aishwarya B C
AI builder | Data Science student | Turning real-world problems into intelligent solutions

Updates

Aishwarya B C started this project — Apr 23, 2026 01:28 PM EDT

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.