Spanning Tree Basis for Unbiased Averaging of Network Topologies

TL;DR

This paper introduces a spectral spanning tree basis to compute unbiased group-level brain network representations, overcoming limitations of traditional averaging and bootstrap methods.

Contribution

It proposes a novel spectral method for averaging maximum spanning trees, reducing bias and computational complexity in group-level brain network analysis.

Findings

The spectral representation accurately captures global properties of group MSTs.

The average tree overlaps with the union of shortest paths in brain networks.

The method improves unbiasedness and efficiency in network comparison.

Abstract

In recent years there has been a paradigm shift from the study of local task-related activation to the organization and functioning of large-scale functional and structural brain networks. However, a long-standing challenge in this large-scale brain network analysis is how to compare network organizations irrespective of their complexity. The maximum spanning tree (MST) has served as a simple, unbiased, standardized representation of complex brain networks and effectively addressed this long-standing challenge. This tree representation, however, has been limited to individual networks. Group-level trees are always constructed from the average network or through a bootstrap procedure. Constructing the group-level tree from the average network introduces bias from individual subjects with outlying connectivities. The bootstrap method can be computationally prohibitive if a good approximation is desired. To address these issues, we propose a novel spectral representation of trees using the spanning tree basis. This spectral representation enables us to compute the average MST and demonstrate that this average tree captures the global properties of all the MSTs in the group and also overlaps with the union of the shortest paths in the functional brain networks.