RP-CNN-Classification

Deep Learning pipeline for Retinitis Pigmentosa (RP) gene group classification using Convolutional Neural Networks and Transfer Learning with ResNet architectures.

---

Overview

Retinitis Pigmentosa (RP) is a heterogeneous group of inherited retinal dystrophies caused by mutations in over 80 genes. This project develops a CNN-based classification pipeline to distinguish between 5 gene groups — both syndromic and non-syndromic — from retinal imaging data, leveraging Transfer Learning with ResNet models pre-trained on ImageNet.

Key Features

• Transfer Learning with ResNet (ResNet-18 / ResNet-50) via PyTorch & torchvision
• 5-class classification covering syndromic and non-syndromic RP gene groups
• K-Fold Cross-Validation for robust model evaluation across data splits
• Detailed metrics: Classification Report, Confusion Matrix, per-class accuracy
• Reproducible training pipeline with configurable hyperparameters

---

Gene Groups

Group	Type	Description
Group 1	Non-syndromic	Autosomal dominant RP genes
Group 2	Non-syndromic	Autosomal recessive RP genes
Group 3	Non-syndromic	X-linked RP genes
Group 4	Syndromic	Usher syndrome genes
Group 5	Syndromic	Bardet-Biedl syndrome genes

---

Project Structure

RP-CNN-Classification/
│
├── notebooks/                  # Jupyter notebooks for experiments
│   └── .gitkeep
│
├── docs/                       # Documentation and figures
│   └── .gitkeep
│
├── .github/
│   └── ISSUE_TEMPLATE/
│       ├── bug_report.md
│       └── feature_request.md
│
├── README.md
├── LICENSE
├── .gitignore
└── requirements.txt

---

Installation

Prerequisites

• Python 3.8+
• CUDA-compatible GPU (recommended)

Setup

# Clone the repository
git clone https://github.com/YOUR_USERNAME/RP-CNN-Classification.git
cd RP-CNN-Classification

# Create and activate a virtual environment
python -m venv venv
source venv/bin/activate        # Linux/macOS
venv\Scripts\activate           # Windows

# Install dependencies
pip install -r requirements.txt

---

Usage

Training

Open and run the main training notebook:

jupyter notebook notebooks/

Cross-Validation

The pipeline uses stratified K-Fold cross-validation. Each fold trains a ResNet model from scratch (with frozen/unfrozen backbone options) and evaluates on the held-out fold.

Evaluation Metrics

After training, the pipeline produces:

• Confusion Matrix — per-fold and aggregated
• Classification Report — precision, recall, F1-score per gene group
• Accuracy curves — training vs. validation per epoch

---

Results

Results section to be updated after final experiments.

Model	Folds	Mean Accuracy	Mean F1 (macro)
ResNet-18	5	—	—
ResNet-50	5	—	—

---

Dependencies

See requirements.txt for the full list. Core libraries:

• PyTorch — model training and inference
• torchvision — ResNet models and image transforms
• scikit-learn — cross-validation and metrics
• Pillow — image loading
• matplotlib / seaborn — visualization

---

Academic Context

This project was developed as part of a Deep Learning coursework focused on medical image analysis and genetic classification of retinal diseases.

---

License

This project is licensed under the MIT License — see the LICENSE file for details.