Learning Clique Subgraphs in Structural Brain Network Classification with Application to Crystallized Cognition

Abstract

Structural brain networks constructed from diffusion MRI are important biomarkers for understanding human brain structure and its relation to cognitive functioning. There is increasing interest in learning differences in structural brain networks between groups of subjects in neuroimaging studies, leading to a variable selection problem in network classification. Traditional methods often use independent edgewise tests or unstructured generalized linear model (GLM) with regularization on vectorized networks to select edges distinguishing the groups, which ignore the network structure and make the results hard to interpret. In this paper, we develop a symmetric bilinear logistic regression (SBLR) with elastic-net penalty to identify a set of small clique subgraphs in network classification. Clique subgraphs, consisting of all the interconnections among a subset of brain regions, have appealing neurological interpretations as they may correspond to some anatomical circuits in the brain related to the outcome. We apply this method to study differences in the structural connectome between adolescents with high and low crystallized cognitive ability, using the crystallized cognition composite score, picture vocabulary and oral reading recognition tests from NIH Toolbox. A few clique subgraphs containing several small sets of brain regions are identified between different levels of functioning, indicating their importance in crystallized cognition.

BrainIB: Interpretable Brain Network-based Psychiatric Diagnosis with Graph Information Bottleneck

“…The identification of predictive subnetworks and edges is an essential procedure for the development of modern psychiatric diagnostic models [28]. Traditionally, this is done by treating functional connectivities as features and performing feature selection to preserve the most salient connections.…”

“…However, most of subgraph discovery model are always based on the node selection such as SIB [24]. Note that, edges (i.e., functional connectivities) are more critical in psychiatric diagnosis [28].…”

Clustered Graph Matching for Label Recovery and Graph Classification

“…One popular method for graph-level classification is to use graph kernels to measure the similarity of graphs and then define a classifier on the similarity matrices, see [11], [46], [54]. Traditional classifiers on vectorized graphs are also equipped with regularizations that enforce some network structure [4], [61], [62]. Deep learning based classifiers are also popular, especially with the growing interest in neural networks; for example [20], [45].…”

Graph Frequency Analysis of COVID-19 Incidence to Identify County-Level Contagion Patterns in the United States

“…Networks are ubiquitous and their graph representations offer an ideal tool to record and analyze massive amounts of data from almost every aspect of human life [1]: social networks [2,3], traffic networks [4,5] and biological networks [6,7], just to name a few. Network data usually reside on irregular structures, requiring graph algorithms for analysis of emergent complex behavior [8].…”