Minimum spanning tree analysis of brain networks: A systematic review of network size effects, sensitivity for neuropsychiatric pathology, and disorder specificity

Abstract

Brain network characteristics’ potential to serve as a neurological and psychiatric pathology biomarker has been hampered by the so-called thresholding problem. The minimum spanning tree (MST) is increasingly applied to overcome this problem. It is yet unknown whether this approach leads to more consistent findings across studies and converging outcomes of either disease-specific biomarkers or transdiagnostic effects. We performed a systematic review on MST analysis in neurophysiological and neuroimaging studies (N = 43) to study consistency of MST-metrics between different network sizes, and assessed disease specificity and transdiagnostic sensitivity of MST-metrics for neurological and psychiatric conditions. Analysis of data from control groups (12 studies) showed that MST leaf fraction but not diameter decreased with increasing network size. Studies showed a broad range in metric values, suggesting that specific processing pipelines affect MST topology. Contradicting findings remain in the inconclusive literature of MST brain network studies, but some trends were seen: 1) a more line-like organization characterizes neurodegenerative disorders across pathologies, and is associated with symptom severity and disease progression; 2) neurophysiological studies in epilepsy show frequency band specific MST alterations that normalize after successful treatment; and 3) less efficient MST topology in alpha band is found across disorders associated with attention impairments.

How Functional Connectivity Measures Affect the Outcomes of MST Neuronal Network Characteristics in Patients with Schizophrenia Compared to Healthy Controls

“…In a current extensive review aimed to find the repeatable constellation of graph-theory networks reconstructions based on MST authors indicated, that regarding the adult psychiatric population, especially schizophrenia, the results are indisputably conflicting. However, the problem of using FC input gained from different connectivity computations was not considered as a possibly important origin of inconsistency [34]. To verify if two different types of FC indicators might generate different or even contradictory results on the extent to which the global neural network of schizophrenia patients differs from the healthy controls, we conducted two identical computational analyses, one with PLI as a synchronization measure, and second with PLV.…”

Persistent Homology-based Functional Connectivity Explains Cognitive Ability: Life-span Study

“…While doing it, at some point, all nodes are connected into one connected component, which is the tree (i.e., no loops) connecting all N nodes with N-1 edges, of which weights are in W 0 (i.e., MST). The studies related to MST have described the 5/20 6/20 structure having MST as the "backbone" of the information flow [4,35,46], and here, we also called this structure "backbone." Also, when edges with weights in W 10 are assigned to the network, additional triangles are created upon the backbone structure, and we called these edges with weights in W 10 "cycle.…”

“…The MST method locates a unique spanning tree that connects N brain regions with N-1 edges at minimum cost (i.e., maximizing synchronization between brain regions). Under the assumption that the brain network is structured as a kind of transportation network, MST will serve as an important backbone of information flow in a weighted brain network [4, 35, 45, 46]. The MST considers topological efficiency (MST has the N-1 shortest path) and strength efficiency (MST has the highest connectivity strength among the possible trees).…”

The longitudinal relation between executive functioning and multilayer network topology in glioma patients

“…The MST recapitulates the network's backbone, by incorporating each node and optimizing the overall weight of the network with n-1 connections (in our case 77 connections) and no cycles. The MST has been amply applied in MEG, with a recent meta-analysis revealing consistent transdiagnostic MST alterations (Blomsma et al, 2022). We set all interlayer link weights to 1 (Figure 1).…”