Brain Networks Flow-Topology via Variance Minimization in the Wasserstein Space

TL;DR

This paper presents a new method combining Hodge decomposition and Wasserstein variance minimization to robustly detect topological differences in weighted networks, especially in noisy brain network data.

Contribution

It introduces a topological variability testing framework that reduces noise sensitivity by analyzing decomposed network signals in Wasserstein space.

Findings

Method suppresses noise-induced perturbations

Maintains sensitivity to genuine topological differences

Effective in analyzing functional brain networks

Abstract

This work introduces a novel framework for testing topological variability in weighted networks by combining Hodge decomposition with Wasserstein variance minimization. Traditional approaches that analyze raw edge weights are susceptible to noise driven perturbations, limiting their ability to detect meaningful structural differences between network populations. Network signals are decomposed into various components using combinatorial Hodge theory, then topological disparity is quantified via the 2-Wasserstein distance between persistence diagrams. The test statistic measures variance reduction when comparing within group to between group dispersions in the Wasserstein space. Simulations demonstrate that the proposed method suppresses small random perturbations while maintaining sensitivity to genuine topological differences, particularly when applied to Hodge decomposed flows rather than raw edge weights. The framework is applied to functional brain networks from the Multimodal Treatment of ADHD dataset, comparing cannabis users and non-users