DBGDGM: Dynamic Brain Graph Deep Generative Model

TL;DR

This paper introduces DBGDGM, a novel deep generative model that captures the temporal evolution of brain connectivity networks from fMRI data, improving graph generation and link prediction while aligning with neuroscience findings.

Contribution

The paper presents a dynamic, unsupervised model for brain graph analysis that models evolving communities and node embeddings over time, addressing limitations of static graph approaches.

Findings

DBGDGM outperforms baselines in graph generation and dynamic link prediction.

The model achieves comparable results in graph classification tasks.

Analysis shows learned communities align with known brain functional connectivity networks.

Abstract

Graphs are a natural representation of brain activity derived from functional magnetic imaging (fMRI) data. It is well known that clusters of anatomical brain regions, known as functional connectivity networks (FCNs), encode temporal relationships which can serve as useful biomarkers for understanding brain function and dysfunction. Previous works, however, ignore the temporal dynamics of the brain and focus on static graphs. In this paper, we propose a dynamic brain graph deep generative model (DBGDGM) which simultaneously clusters brain regions into temporally evolving communities and learns dynamic unsupervised node embeddings. Specifically, DBGDGM represents brain graph nodes as embeddings sampled from a distribution over communities that evolve over time. We parameterise this community distribution using neural networks that learn from subject and node embeddings as well as past community assignments. Experiments demonstrate DBGDGM outperforms baselines in graph generation, dynamic link prediction, and is comparable for graph classification. Finally, an analysis of the learnt community distributions reveals overlap with known FCNs reported in neuroscience literature.