Bayesian network analysis revealed the connectivity difference of the default mode network from the resting-state to task-state

Xia Wu, Xinyu Yu, Li Yao, Rui LiFrontiers Comput. Neurosci 2014 · OpenReview

TL;DR

This study uses Bayesian Networks to map the directed connections within the brain's Default Mode Network (DMN) during both resting and task states. While the overall structure of the DMN remains stable, the way information flows between its regions significantly changes between resting and performing a semantic judgment task, suggesting a dynamic information processing mechanism.

What Problem Does This Paper Address?

The Default Mode Network (DMN) is a well-established set of brain regions that are highly active when our minds wander or we're not focused on external tasks, but become less active during demanding cognitive tasks. While we know which regions are part of the DMN and that they're interconnected, understanding how information flows between these regions (effective connectivity) and how this flow changes between different mental states (resting vs. task) has been a significant challenge. Previous studies often focused on undirected functional connectivity or used methods that didn't capture the directed nature of information transfer. This paper aims to fill this gap by using a Bayesian Network approach to model the directed effective connectivity of the DMN and compare it across resting and task states.

Methods

The researchers employed a Bayesian Network (BN) analysis, a powerful statistical tool for modeling probabilistic relationships between variables. In this context, the nodes of the BN represent specific brain regions within the DMN, and the directed arcs represent the effective connectivity (information flow) between them. The BN approach is advantageous because it can learn the global connectivity pattern without requiring prior assumptions about the network structure and can represent directed dependencies.

Here's a breakdown of their methodology:

  • Data Acquisition: fMRI data was collected from 14 healthy volunteers during two conditions: resting-state (eyes closed, no specific task) and a semantic judgment task (judging if visually presented words were dangerous).
  • Preprocessing: Standard fMRI preprocessing steps were applied, including realignment, normalization, smoothing, and filtering to remove noise and artifacts.
  • Region of Interest (ROI) Identification: Group Independent Component Analysis (ICA) was used to identify the DMN in both resting and task states. From these DMN maps, eight specific regions were defined as Regions of Interest (ROIs) for the BN analysis. These ROIs were based on local maxima of functional connectivity.
  • Bayesian Network Learning: For each subject and each condition, the time series data from the ROIs were extracted. A linear Gaussian Bayesian Network model was then learned to represent the directed effective connectivity. The BIC (Bayesian Information Criterion) score was used to select the optimal network structure.
  • Statistical Testing: To compare the connectivity between the resting and task states, a randomized permutation test was performed. This involved comparing the BN connectivity weights between the two groups and assessing the significance of any differences by repeatedly shuffling group assignments.

Datasets

The study utilized fMRI data collected from 14 healthy volunteers. This data was acquired during two distinct conditions: resting-state and a semantic judgment task. No new datasets were introduced or made publicly available as part of this publication.

Key Insights and Ideas

  • Stable DMN Architecture, Dynamic Flow: The most striking finding is that the DMN's overall organizational structure (which regions are connected to which) remains remarkably stable between resting and task states. However, the strength and direction of these connections change significantly, indicating that the brain reconfigures how it processes information within the DMN depending on the cognitive demands.
  • Hubs and Non-Hubs: The DMN can be conceptually divided into "hub" regions (PCC, MPFC, IPC) and "non-hub" regions (ITC, HC). The study reinforces the idea that non-hub regions primarily send information to the hub regions, which then integrate this information. This "from non-hub to hub" flow is consistent across both states.
  • PCC as a Confluent Node: The Posterior Cingulate Cortex (PCC) consistently acts as a central integration point, receiving input from multiple other DMN regions in both resting and task states. This highlights its crucial role in coordinating DMN activity.
  • Semantic Task Influence: The semantic judgment task specifically led to increased connectivity involving the lateral temporal regions (ITC), suggesting their direct involvement in processing semantic information within the DMN framework.

Key Results

  • The Bayesian Network analysis revealed a consistent "from non-hub subsystem to hub system" organizational pattern in the DMN for both resting and task states.
  • Significant differences in connection weights were observed between the resting and task states. For example, connections from the lateral temporal cortex (ITC) to other DMN regions showed increased strength during the task state, aligning with the task's semantic nature.
  • While the overall structure was stable, specific connections showed notable changes in weight, indicating altered information processing mechanisms. The paper highlights 12 connections with increased weights during the task state, with nine directly involving the ITC.

Limitations and Future Work

The authors acknowledge several limitations:

  • Acyclicity: Bayesian Networks, by definition, are Directed Acyclic Graphs (DAGs), meaning they cannot model reciprocal (bidirectional) connections or self-influencing connections. This limits the ability to fully capture the complex, potentially reciprocal nature of brain connectivity.
  • Snapshot of Dynamics: The BN model provides a snapshot of dynamic processes and doesn't explicitly disclose temporal causal relations between nodes.
  • Group-Level Analysis: The effective connectivity was estimated at the group level, making it difficult to correlate specific connectivity patterns with individual behavioral performance.

Future work suggested includes:

  • Utilizing dynamic Bayesian Networks to capture temporal interrelationships and model reciprocal connections.
  • Exploring the relationship between DMN connectivity and behavioral performance to gain a deeper understanding of the DMN's role in cognition.
  • Investigating how these connectivity patterns relate to neurological or psychiatric disorders, building on previous work in this area.

Summary updated

Abstract

Functional magnetic resonance imaging (fMRI) studies have converged to reveal the default mode network (DMN), a constellation of regions that display co-activation during resting-state but co-deactivation during attention-demanding tasks in the brain. Here, we employed a Bayesian network (BN) analysis method to construct a directed effective connectivity model of the DMN and compared the organizational architecture and interregional directed connections under both resting-state and task-state. The analysis results indicated that the DMN was consistently organized into two closely interacting subsystems in both resting-state and task-state. The directed connections between DMN regions, however, changed significantly from the resting-state to task-state condition. The results suggest that the DMN intrinsically maintains a relatively stable structure whether at rest or performing tasks but has different information processing mechanisms under varied states.
BibTeX
@inproceedings{wu2014bayesian,
  author = {Xia Wu and Xinyu Yu and Li Yao and Rui Li},
  title = {{Bayesian network analysis revealed the connectivity difference of the default mode network from the resting-state to task-state}},
  booktitle = {Frontiers Comput. Neurosci},
  year = {2014},
  url = {https://openreview.net/forum?id=QAApVhjUTNc},
}
Figure