RetinalScope
Medical Vision Analytics

Explainable AI model for early detection of retinal diseases using deep learning. Helping healthcare professionals diagnose faster with transparent, interpretable results.

7
Diseases Detected
3,285
Training Images
91%+
Accuracy
AI Analysis Active Diabetic Retinopathy: 87% Macular Edema: 23%

The Problem We're Solving

Early diagnosis saves vision, but traditional methods have limitations

Time-Consuming

Manual diagnosis requires extensive time from skilled ophthalmologists

Inter-Observer Variability

Different specialists may reach different conclusions for the same case

Irreversible Vision Loss

Late diagnosis can lead to permanent vision damage

Our Solution

An automated and explainable AI system that assists healthcare professionals in detecting retinal anomalies accurately and interpretably, making diagnosis faster, more transparent, and educational.

Key Objectives:

  • Detect multiple retinal diseases from fundus images
  • Explain predictions in plain English
  • Use Grad-CAM heatmaps for visual explanation

Targeted Diseases:

• Opacity
• Diabetic Retinopathy
• Glaucoma
• Macular Edema
• Macular Degeneration
• Retinal Vascular Occlusion
• Normal Eye

Key Features

Advanced AI capabilities for comprehensive retinal analysis

Multi-Label Classification

Simultaneously detects multiple retinal conditions with confidence scores using ResNet50 transfer learning.

Explainable AI

GPT-4 integration converts technical predictions into human-readable medical explanations.

Grad-CAM Visualization

Visual heatmaps show exactly which areas of the retina influenced the AI's decision.

Image Upload Interface

Easy-to-use web interface for uploading fundus images and receiving instant analysis.

Real-time Processing

Fast inference with image validation, preprocessing, and disease prediction in seconds.

Comprehensive Dataset

Trained on 3,285 fundus images from Kaggle and public medical datasets.

Technology Stack

Built with cutting-edge AI and web technologies

Core Technologies

PT
PyTorch & Torchvision
Deep learning model training and computer vision
AI
OpenAI GPT-4
Natural language explanations and medical insights
ST
Streamlit
Interactive web interface and inference server
NP
NumPy
Numerical computing and data handling

Model Architecture

Base Model

ResNet50 with transfer learning

Modification

4th layer retrained for retinal disease classification

Output

Multi-label classification for 7 conditions

Explainability

Grad-CAM heatmaps + GPT-4 explanations

Dataset Details

Total Images
3,285 fundus images
Abnormal Cases
3,210 images
Normal Cases
575 images
Sources
Kaggle + EyePACS

Sample Output & Results

See how our AI explains its diagnostic decisions

Example Analysis

Detected Conditions:

Diabetic Retinopathy 87%
Macular Edema 23%
Glaucoma 8%

AI Explanation:

"The analysis reveals significant signs of diabetic retinopathy with 87% confidence. Key indicators include microaneurysms, hemorrhages, and exudates visible in the retinal blood vessels. The Grad-CAM heatmap highlights areas of concern around the optic disc and macula. Early intervention is recommended to prevent progression to proliferative diabetic retinopathy."

Grad-CAM Visualization

High Attention
Medium Attention
Low Attention

Get Started

Easy installation and setup process

1

Clone Repository

Download the project from GitHub

2

Install Dependencies

Set up Python environment and packages

3

Run Application

Start the web interface

1. Clone the Repository

git clone https://github.com/username/RetinalAnomalyDetection.git
cd RetinalAnomalyDetection

2. Install Requirements

# Python Version: 3.12.3
pip install -r requirements.txt

3. Run the Application

python main.py

Requirements

  • • Python 3.12.3 or higher
  • • PyTorch and Torchvision
  • • OpenAI API key (for explanations)
  • • Streamlit for web interface
  • • NumPy for data processing

Ready to Try RetinalScope?

Experience the future of retinal disease detection with explainable AI

Open source • Free to use • Educational purpose