Brain connectivity matrix super-resolution using graph neural networks

Group project at Imperial College London as part of the Deep Graph-based Learning (DGL) module by Prof. Islem Rekik.

📄 View report

This project focuses on using generative Graph Neural Networks (GNNs) to predict high-resolution (HR) brain connectivity graphs from their low-resolution (LR) counterparts. This task is motivated by the need to enhance the resolution of brain connectivity data, which is crucial for advancing our understanding of brain function. The problem involves learning a mapping function f that can accurately transform a low-resolution brain matrix ALR into a high-resolution brain matrix AHR. This task is essential because high-resolution brain graphs provide more detailed and precise information about neural connections, which can significantly improve the analysis and diagnosis of neurological conditions. By developing a model that can infer high-resolution connectivity from low-resolution data, we aim to overcome the limitations of current imaging techniques and pave the way for more accurate and insightful brain research.

(Isallari & Rekik, 2021) (Isallari & Rekik, 2020) (Gao & Ji, 2019) (Dwivedi & Bresson, 2021) (Gao & Ji, 2019) (Hamilton et al., 2017) (Xu et al., 2018) (Liu et al., 2017) (Yun et al., 2019)

Diagram of the model architecture used in the project, showcasing the generative Graph Neural Network (GNN) approach to predict high-resolution brain connectivity graphs from low-resolution inputs. We used a modified version of GraphUNet for upsampling the low-resolution brain connectivity matrix.
denisfouchard/fMRI-connectivity-super-resolution denisfouchard/fMRI-connectivity-super-resolution

References

2021

  1. Brain Graph Super-Resolution Using Adversarial Graph Neural Network with Application to Functional Brain Connectivity
    Megi Isallari and Islem Rekik
    2021
  2. A Generalization of Transformer Networks to Graphs
    Vijay Prakash Dwivedi and Xavier Bresson
    AAAI Workshop on Deep Learning on Graphs: Methods and Applications, 2021

2020

  1. Graph Super-Resolution Network for predicting high-resolution connectomes from low-resolution connectomes
    Megi Isallari and Islem Rekik
    In International Workshop on PRedictive Intelligence In MEdicine, 2020

2019

  1. Graph U-Nets
    Hongyang Gao and Shuiwang Ji
    In International Conference on Machine Learning, 2019
  2. Graph Transformer Networks
    Seongjun Yun, Minbyul Jeong, Raehyun Kim, and 2 more authors
    CoRR, 2019

2018

  1. How Powerful are Graph Neural Networks?
    Keyulu Xu, Weihua Hu, Jure Leskovec, and 1 more author
    CoRR, 2018

2017

  1. Inductive representation learning on large graphs
    Will Hamilton, Zhitao Ying, and Jure Leskovec
    Advances in neural information processing systems, 2017
  2. Longitudinal test-retest neuroimaging data from healthy young adults in southwest China
    Wei Liu, Dongtao Wei, Qunlin Chen, and 7 more authors
    Scientific data, 2017