RetiSpec 👁️

Explainable AI-Based Diabetic Retinopathy Detection System

📌 Overview

RetiSpec is an AI/ML project developed for detecting and classifying Diabetic Retinopathy (DR) from retinal fundus images using deep learning and Explainable AI techniques.

The model classifies retinal images into 5 severity stages:

• No DR
• Mild
• Moderate
• Severe
• Proliferative DR

The project uses ConvNeXt-Tiny as the backbone architecture along with preprocessing, threshold optimization, Test-Time Augmentation (TTA), and Explainable AI visualizations.

---

📂 Datasets Used

The model was trained using multiple publicly available retinal fundus datasets:

• APTOS 2019 Blindness Detection
• Messidor-2
• EyePACS

These datasets were combined to improve model generalization and robustness across different retinal image distributions.

---

🚀 Features

• Deep learning-based DR classification
• ConvNeXt-Tiny architecture
• Ben Graham retinal preprocessing
• Test-Time Augmentation (TTA)
• Threshold optimization
• Explainable AI visualizations
• Flask web application
• Interactive frontend using HTML, CSS, and JavaScript

---

📊 Performance

• Accuracy: 93%
• Quadratic Weighted Kappa Score: 0.81

---

🔥 Explainable AI (XAI)

The project includes Explainable AI techniques to improve interpretability and transparency of predictions.

Techniques Used

• Grad-CAM++
• Gradient × Input visualization

These visualizations help identify important retinal regions influencing the model prediction.

---

🛠️ Tech Stack

Backend

• Python
• Flask
• PyTorch

Frontend

• HTML
• CSS
• JavaScript

Libraries

• OpenCV
• Albumentations
• timm
• torchvision
• NumPy
• Pillow

---

📁 Project Structure

RetiSpec/
│
├── static/
│   ├── css/
│   └── js/
│
├── templates/
│   └── index.html
│
├── app.py
├── config.py
├── dataset.py
├── evaluate.py
├── model.py
├── preprocessing.py
├── requirements.txt
├── threshold_optimization.py
├── train.py
├── tta.py
├── utils.py
├── xai_gradcam.py
├── xai_shap.py

---

▶️ Run the Project

Install dependencies:

pip install -r requirements.txt

Run the Flask server:

python app.py

---

📸 Outputs

The repository includes:

• prediction outputs
• Grad-CAM visualizations
• frontend screenshots

---

👩‍💻 Developed By

• Dena D
• Jeffisha Jemi J

Final Year AI/ML Engineering Project

---

⚠️ Note

This project is developed for educational and research purposes only.