Aligned Spectral Embedding for Brain Surface Analysis

This repository contains the python implementation of the graph spectral alignment. We perform this spectral alignment to overcome the limitations of spectral graph convolution networks in our work "Graph Convolutions on Spectral Embeddings for Cortical Surface Parcellation", published in Medical Image Analysis, January 2019.

What does the reopositery do?

main.py
1. FreeSurfer brain surfaces is read from the "dataset" folder.
2. Spectral embedding of the brain graph is computed.
3. Spectral basis of each subject is aligned to a common reference from the dataset.
4. Computed embeddings, transformation and aligned spectral embeddings are saved in "output" folder in ".pt" for DeepLearning algorithms later use in PyTorch.

Where to find the dataset?

The MindBoggle brain surfaces dataset is available to download here.
The FreeSurfer processed ADNI surfaces dataset is available to download here.

Package Requirements

math
matplotlib
mne
nibabel
numpy
python3, pytorch>1.0
pickle
pandas
scipy
time, timeit
vtk

Installation

To use the requirements.txt file to create an identical environment on the same machine or another machine:

conda create --name myenv --file requirements_linux.txt

To use the requirements.txt file to install its listed packages into an existing environment:

conda install --name myenv --file requirements_linux.txt

Usage

python3 main.py

REFERENCE

Please cite our papers if you use this code in your own work:

@article{gopinath2019graph,
  title={Graph convolutions on spectral embeddings for cortical surface parcellation},
  author={Gopinath, Karthik and Desrosiers, Christian and Lombaert, Herve},
  journal={Medical image analysis},
  volume={54},
  pages={297--305},
  year={2019},
  publisher={Elsevier}
}