GRN, a graph registration network as a plug-in to conventional GNN classifiers to register graphs into a fixed connectional template for boosting the predictive power of GNN classifiers, coded up in Python by Zeynep Gürler and Mohammed Amine Gharsallaoui. Please contact [email protected] for inquiries. Thanks.
Graph registration network for boosting diagnosis of neurological disorders
Zeynep Gürler1, Mohammed Amine Gharsallaoui1,2, Islem Rekik1 1BASIRA Lab, Faculty of Computer and Informatics, Istanbul Technical University, Istanbul, Turkey 2National School for Computer Science (ENSI), Mannouba, Tunisia
Abstract: *Deep learning has gained critical importance in the computer-aided diagnosis of neurological disorders. Formerly published methods worked on the data that is driven from medical imaging tools overlooking the complex and internally connected structure of the brain. Brain graphs are excellent representations to explore the biological roadmaps of the human brain. Although there are several conventional graph neural networks (GNNs) for graph classification, these GNNs fail to preserve the complexity and richness of brain connectivity graphs. Therefore, we propose an integrational graph classification framework that is composed of two components respectively, (1) graph registration network (GRN) and (2) a conventional plug-in graph neural network (GNN) to boost diagnosis of neurological disorders. Our GRN is a graph generative adversarial network (gGAN), which registers brain graphs into a fixed brain population template, namely a connectional brain template (CBT) (i.e., a brain template that captures the most common features across a brain graph population) to capture their most discriminative features with respect to the population center while holding their connectional and structural information. Our GRN can be integrated into any GNN working in an end-to-end fashion to boost its prediction accuracy. Our experiments showed that GRN remarkably boosted the prediction accuracy of conventional GNN models.
Our framework is a brain graph classification framework based on a graph registration architecture comprising a registrator network which learns how to register brain graphs into a fixed connectional brain template (CBT). Our learning-based framework comprises two key steps. (1) Registration of input graphs into the population CBT, (2) Classification of brain graphs with the plug-in GNN using the registered versions of the brain graphs. Experimental results against comparison methods demonstrate that our framework can achieve the best results in terms of mean prediction accuracy. We evaluated GRN on two datasets, where each subject is represented by a morphological brain graph. The first dataset (ASD/NC dataset) is collected from the Autism Brain Imaging Data Exchange ABIDE I public dataset (http://fcon_1000.projects.nitrc.org/indi/abide/). The second dataset (LMCI/AD dataset) is collected from Alzheimer’s Disease Neuroimaging Initiative (ADNI) database GO public dataset (adni.loni.usc.edu).
More details can be found at: (link to the paper) and our research paper video on the BASIRA Lab YouTube channel (link).
- Python 3.8
- PyTorch 1.7.0
- Torch-geometric
- Torch-sparse
- Torch-scatter
- Scikit-learn 0.23.0+
- Matplotlib 3.1.3+
- Numpy 1.20.1+
GRN is coded in Python 3.8 on Windows 10. GPU is not needed to run the code. This code has been slightly modified to be compatible across all PyTorch versions. main.py is the implementation of the brain graph classification framework proposed by Graph registration network for boosting diagnosis of neurological disorders paper. In this repo, we release the GRN source code trained and tested on a simulated data as shown below:
Data preparation
We provide a demo code for the usage of GRN for brain graph registration to boost the predictive performance of GNN classifiers. In main.py, we train GRN on a simulated dataset with 240 subjects which are pre-obtained using dataset.py. If you need to generate new simulated data with different distribution, you can change the mean correlation matrix and connectivity mean values in dataset.py.
If you want to train GRN on pre-obtained simulated data, you can set the GNN type and the model parameters in arg_parse function of demo.py and execute,
python demo.py
To run GRN, we generate a fixed connectional brain template using netNorm: https://github.com/basiralab/netNorm-PY.
demo.py generates the following outputs:
Network Normalization for Integrating Multi-view Networks (netNorm): Dhifallah, S., Rekik, I., 2020, Estimation of connectional brain templates using selective multi-view network normalization
To add.
@article{gurler2022,
title={Graph registration network for boosting diagnosis of neurological disorders},
author={Gurler Zeynep, Gharsallaoui Mohamed Amine, Rekik Islem},
journal={Computerized Medical Imaging and Graphics},
volume={},
pages={},
year={2022},
publisher={Elsevier}
}