Path analysis: A method to estimate altered pathways in time-varying graphs of neuroimaging data

Haleh Falakshahi, Hooman Rokham, Zening Fu, Armin Iraji, Daniel H. Mathalon, Judith M. Ford, Bryon A. Mueller, Adrian Preda, Theo G.m. van Erp, Jessica A. Turner, Sergey Plis, Vince D. CalhounNetwork 2022 · DOI

TL;DR

This paper introduces a novel "path analysis" method to better understand brain network disruptions in schizophrenia by looking at how entire pathways, not just individual connections, are altered over time. By analyzing these multi-step brain routes, the researchers identified specific "disconnectors" and "connectors" in the brains of individuals with schizophrenia, offering a more nuanced view of the disorder's impact on brain function.

What Problem Does This Paper Address?

Traditional graph-theoretical approaches to studying brain networks, particularly in conditions like schizophrenia, often focus on individual connections (edges) between brain regions (nodes) or on overall network properties. While useful, this overlooks the crucial information encoded in the sequences of connections that form pathways. These multi-step pathways are thought to be vital for complex cognitive functions, and their disruption could be a key factor in psychiatric disorders. Existing methods struggle to pinpoint how these longer routes are altered, making it difficult to fully grasp the mechanisms of conditions like schizophrenia, which are characterized by widespread functional disconnectivity or abnormal integration. This paper aims to fill that gap by developing a method to analyze these altered pathways in dynamic brain networks.

Methods

The core of this paper is a novel "path analysis" algorithm designed to identify altered pathways in brain graphs derived from neuroimaging data. The method involves several key steps:

  • Graph Estimation: They use Gaussian Graphical Models (GGMs) to represent brain networks, estimating partial correlation matrices to define connections (edges) between intrinsic connectivity networks (ICNs, the nodes). This is done for both static (whole-scan) and dynamic (time-varying) functional connectivity.
  • Dynamic Network Analysis: For dynamic analysis, they employ a sliding window approach to capture time-varying functional connectivity and then use k-means clustering to identify distinct "brain states."
  • Path Analysis Algorithm: The novel algorithm identifies two main types of alterations:
  • Disconnection: Detects pathways present in control groups but absent in patient groups. This is achieved by identifying edges that, if removed in the patient group, break connections within previously connected components.
  • Abnormal Integration: Identifies pathways absent in control groups but present in patient groups. This is done by finding new edges that create novel paths.
  • Covariance Decomposition: For cases where paths exist in both control and patient groups, they use a covariance decomposition method to quantify the contribution of each path to the overall correlation between two brain regions. This helps understand how different path trajectories contribute to brain function.

Datasets

The study utilized resting-state fMRI data from two groups:

  • 160 healthy controls (HC)
  • 151 individuals with schizophrenia (SZ), matched for age, gender, and mean framewise displacement.

The data was collected from seven different sites as part of the Function Biomedical Informatics Research Network (fBIRN). The authors also mention using data from the Human Connectome Project (HCP) and Genomics Superstruct Project (GSP) to generate reproducible ICN templates for their analysis.

Key Insights and Ideas

  • Beyond Pairwise Connections: The most significant insight is that focusing solely on direct connections between brain regions misses crucial information about how these regions communicate through multi-step pathways.
  • Pathways as Biomarkers: Alterations in these pathways, rather than just individual edge changes, can serve as more sensitive biomarkers for psychiatric disorders like schizophrenia.
  • Dynamic Brain States Matter: Brain connectivity is not static; different "brain states" reveal distinct patterns of pathway alterations, highlighting the importance of dynamic analysis.
  • Compensatory Mechanisms: The identification of new edges creating additional paths in schizophrenia patients might suggest compensatory mechanisms attempting to maintain function despite underlying disruptions.

Key Results

  • Static Analysis: Identified 6 "disconnectors" and 2 "connectors" in static brain graphs, indicating missing or added pathways within and between functional domains, particularly in the visual, subcortical, cognitive control, and auditory networks.
  • Dynamic Analysis: Across five identified brain states, they found a total of 14 disconnectors and 16 connectors. State 2, characterized as a "weak-connected state," showed significant alterations.
  • Specific Network Findings: Disrupted pathways were frequently observed within the default mode network and cognitive control networks in schizophrenia patients.
  • Covariance Decomposition: In cases where paths existed in both groups, the analysis revealed that these paths often took unique trajectories in schizophrenia patients, contributing significantly to the overall correlation between brain regions. For example, in State 3, new paths in SZ patients accounted for a substantial portion of the correlation between specific brain regions.

Limitations and Future Work

  • Limitations:
    • The study focused on undirected graphs, which might miss nuances present in directed connectivity.
  • The analysis was primarily applied to resting-state fMRI data, and future work could explore task-based fMRI or other neuroimaging modalities.
  • Future Work:
    • Expand the approach to include directed network analysis for a more comprehensive understanding of brain connectivity.
  • Replicate the findings in additional datasets to confirm their generalizability.
  • Investigate the potential compensatory role of identified "connectors" in schizophrenia.
  • Apply the method to other neurological and psychiatric conditions characterized by disconnectivity or abnormal integration.

Summary updated

Abstract

Graph-theoretical methods have been widely used to study human brain networks in psychiatric disorders. However, the focus has primarily been on global graphic metrics with little attention to the information contained in paths connecting brain regions. Details of disruption of these paths may be highly informative for understanding disease mechanisms. To detect the absence or addition of multistep paths in the patient group, we provide an algorithm estimating edges that contribute to these paths with reference to the control group. We next examine where pairs of nodes were connected through paths in both groups by using a covariance decomposition method. We apply our method to study resting-state fMRI data in schizophrenia versus controls. Results show several disconnectors in schizophrenia within and between functional domains, particularly within the default mode and cognitive control networks. Additionally, we identify new edges generating additional paths. Moreover, although paths exist in both groups, these paths take unique trajectories and have a significant contribution to the decomposition. The proposed path analysis provides a way to characterize individuals by evaluating changes in paths, rather than just focusing on the pairwise relationships. Our results show promise for identifying path-based metrics in neuroimaging data.
BibTeX
@inproceedings{falakshahi2022path,
  author = {Haleh Falakshahi and Hooman Rokham and Zening Fu and Armin Iraji and Daniel H. Mathalon and Judith M. Ford and Bryon A. Mueller and Adrian Preda and Theo G.M. van Erp and Jessica A. Turner and Sergey Plis and Vince D. Calhoun},
  title = {{Path analysis: A method to estimate altered pathways in time-varying graphs of neuroimaging data}},
  booktitle = {Network},
  year = {2022},
  doi = {10.1162/netn\_a\_00247},
}
Figure