AI-Enabled Early Detection of Diabetic Retinopathy Using Fundus Images

Model Overview

Model Name: Diabetic Retinopathy Detector
Version: 1.0
Type: Image Classification
Framework: TensorFlow / Keras
ReadyTensor ID: To be auto-filled upon upload

---

Problem Statement

Diabetic Retinopathy (DR) is a major cause of vision impairment in diabetic patients. Early detection plays a crucial role in enabling timely treatment and preventing irreversible vision loss. This AI model analyzes retinal fundus images to detect early signs of diabetic retinopathy, providing a scalable, non-invasive, and efficient screening solution.

---

Input Features

• fundus_image (Image): RGB retinal fundus photograph
  The model expects input images in standard formats such as JPG or PNG, with proper resizing and normalization.

---

Target Variable

• dr_stage (Multi-class Classification)
  Classes:
  • 0: No DR
  • 1: Mild
  • 2: Moderate
  • 3: Severe
  • 4: Proliferative DR

---

Performance Metrics

Metrics evaluated on a held-out test set comprising 3,000 fundus images.

---

Training Details

• Dataset: APTOS 2019 Blindness Detection (available at Kaggle)
• Training Data Size: Approximately 29,000 labeled images
• Validation Split: 80% training / 20% validation
• Image Size: Resized to 224x224 pixels
• Augmentations: Random rotation, flipping, brightness and contrast adjustments
• Training Duration: Approximately 3 hours on GPU
• Hardware: NVIDIA Tesla V100 (16 GB)

---

Model Artifacts

• dr_model.h5 (Keras model file)
• preprocessing.py (script for image resizing and normalization)
• labels_map.json (mapping between class indices and labels)
• requirements.txt (includes TensorFlow 2.x, OpenCV, NumPy)

---

Usage Instructions

1. Input Format:
   Upload or pass a base64-encoded RGB fundus image.

   Example API input:

   {
     "image": "<base64_encoded_image>"
   }