Topology-Aware Graph Augmentation for Predicting Clinical Trajectories in Neurocognitive Disorders

TL;DR

This paper introduces a topology-aware graph augmentation framework for brain network analysis from fMRI data, improving the generalizability of models in neurocognitive disorder prediction by preserving critical network topology during augmentation.

Contribution

It proposes novel topology-aware augmentation strategies for graph contrastive learning tailored to brain networks, enhancing model robustness and performance.

Findings

TGA outperforms state-of-the-art methods on 1,688 fMRI scans.

Topology-aware augmentations improve model generalizability.

Preserving brain hub regions enhances downstream task accuracy.

Abstract

Brain networks/graphs derived from resting-state functional MRI (fMRI) help study underlying pathophysiology of neurocognitive disorders by measuring neuronal activities in the brain. Some studies utilize learning-based methods for brain network analysis, but typically suffer from low model generalizability caused by scarce labeled fMRI data. As a notable self-supervised strategy, graph contrastive learning helps leverage auxiliary unlabeled data. But existing methods generally arbitrarily perturb graph nodes/edges to generate augmented graphs, without considering essential topology information of brain networks. To this end, we propose a topology-aware graph augmentation (TGA) framework, comprising a pretext model to train a generalizable encoder on large-scale unlabeled fMRI cohorts and a task-specific model to perform downstream tasks on a small target dataset. In the pretext model, we design two novel topology-aware graph augmentation strategies: (1) hub-preserving node dropping that prioritizes preserving brain hub regions according to node importance, and (2) weight-dependent edge removing that focuses on keeping important functional connectivities based on edge weights. Experiments on 1, 688 fMRI scans suggest that TGA outperforms several state-of-the-art methods.