Alzheimer's Detection using Vision Transformers

This project implements an Alzheimer's disease detection system using Vision Transformers (ViT) and 3D Convolutional Neural Networks. It leverages transfer learning techniques to adapt pre-trained models for MRI scan analysis.

Trying the 🤗Hugging Face Demo: Alzheimer Classifier

Table of Contents

• Project Overview
• Installation
• Project Structure
• Usage
• Models
• Data
• Training
• Evaluation
• Testing
• Contributing
• License

Project Overview

This project aims to detect Alzheimer's disease using MRI scans. It implements and compares three different deep learning architectures:

1. 2D Vision Transformer (ViT)
2. 3D Vision Transformer
3. 3D Convolutional Neural Network (CNN)

The project utilizes transfer learning techniques, starting with pre-trained models and adapting them for the specific task of Alzheimer's detection.

Installation

1. Clone the repository:

   git clone https://github.com/yourusername/alzheimers-detection-vit.git
   cd alzheimers-detection-vit
   

2. Create a virtual environment (optional but recommended):

   python -m venv venv
   source venv/bin/activate  # On Windows, use `venv\Scripts\activate`
   

3. Install the required packages:

   pip install -r requirements.txt
   

Project Structure

alzheimer_vit/
├── data/
│   ├── test/
│   └── train/
├── experiments/
├── notebooks/
├── src/
│   ├── data/
│   │   └── data_loader.py
│   ├── models/
│   │   ├── architectures/
│   │   │   ├── vit2d.py
│   │   │   ├── vit3d.py
│   │   │   └── cnn3d.py
│   │   ├── evaluate.py
│   │   └── train.py
│   └── utils/
│       └── logger.py
├── tests/
│   ├── test_dataset.py
│   ├── test_models.py
│   └── test_train.py
├── .env
├── .gitignore
├── config.yaml
├── main.py
├── README.md
└── requirements.txt

Project Visualisation

Classifier Performance (table)

Usage

To run the main script:

python main.py

This will load the dataset, prepare the data, create the specified model, train it, and evaluate its performance.

Models

The project implements three types of models:

1. 2D Vision Transformer (8/16/32 Patch Embeddings) (src/models/architectures/vit2d.py)
2. 3D Vision Transformer (8/16/32 Patch Embeddings) (src/models/architectures/vit3d.py)
3. 3D CNN (ResNet101)(src/models/architectures/cnn3d.py)

You can specify which model to use in the config.yaml file.

Dataset

The project uses the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. Data is loaded from Hugging Face datasets and preprocessed using MONAI transforms.

To start training, run:

python main.py

Training parameters can be adjusted in the config.yaml file.

Evaluation

Model evaluation is performed using the evaluate_model function in src/models/evaluate.py. It uses MONAI's SupervisedEvaluator and calculates the ROC AUC metric.

Contributing

Contributions to this project are welcome. Please follow these steps:

1. Fork the repository
2. Create a new branch (git checkout -b feature/AmazingFeature)
3. Make your changes
4. Commit your changes (git commit -m 'Add some AmazingFeature')
5. Push to the branch (git push origin feature/AmazingFeature)
6. Open a Pull Request

License

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