MRI Classifiers

Introduction

This repository is dedicated to implementing and comparing various machine learning classifiers on MRI images for Alzheimer's disease detection. Our approach includes classical classifiers like Support Vector Machines (SVM) and Random Forest (RF), as well as advanced models such as vision transformers and ResNet-18. We also explore the impact of traditional and novel data augmentation methods, including sharpening, blurring, random erasing, and the use of diffusion models, on classifier performance.

Requirements

To set up the environment, open CMD in Windows or Terminal in Linux, and execute the following command:

pip install -r requirements.txt

Also for data files you need to install 'git LFS'.

Data Augmentation Using Diffusion Models

Training Denoising Diffusion Probabilistic Models (DDPMs)

To train DDPMs for each MRI class, execute the command below for each class:

python src/train_unconditional.py --class_MRI <class mri> --num_epochs 400

Replace <class mri> with one of the following classes:

• MildDemented
• ModerateDemented
• NonDemented
• VeryMildDemented

Training individual models for each class stores them in the diffusion_models directory. All four models are required for data augmentation using DDPMs.

Generating Data Using DDPMs

To generate augmented data:

python src/data_generation.py --number_of_images 5000

This creates a generated folder in the data directory, containing 5000 images for each class.

Configurations

Configuration adjustments can be made in the config.yaml file in the configs folder. For different augmentation methods:

• Classical methods:
  • do_augmentation: true
  • do_generation: false
• Using DDPMs:
  • do_augmentation: true
  • do_generation: true

Execution

To run the project and store results in the result folder, use:

python src/main.py

Docker

If you want to use docker automatically:

• to build:

cd docker
docker build -t env_image .

• to run:

cd ..
docker run -d -it --name container_runner -v .:/app env_image tail -f /dev/null
docker exec -it container_runner bash
pip install -r requirements.txt

Now in the docker container you can do the following things:

• The rest is the same as above.