Unsupervised Representation Learning of Cingulate Cortical Folding Patterns

Official Pytorch implementation for Unsupervised Learning and Cortical Folding paper. The project aims to study cortical folding patterns in the cingulae region thanks to unsupervised deep learning methods.

Installation

We first install the singularity brainvisa image following https://brainvisa.info/web/download.html

We then enter the brainvisa environment:

bv opengl=nv bash

We can use a virtual environment:

virtualenv --python=python3 --system-site-packages venv
. venv/bin/activate

We then download and install the present deep learning package:

git clone https://github.com/neurospin-projects/2021_jchavas_lguillon_deepcingulate
cd 2021_jchavas_lguillon_deepcingulate
pip3 install -e .

As an alternative to the last step, we can also use the provided requirements file:

pip3 install -r requirements.txt

Training the models

Data are available in the data directory.

To train and evaluate each model, we follow the corresponding README:

• README of beta-VAE model
• README of SimCLR model

Results

Dependencies

• python >= 3.6
• pytorch >= 1.4.0
• numpy >= 1.16.6
• pandas >= 0.23.3