Graph Independent Component Analysis Reveals Repertoires of Intrinsic Network Components in the Human Brain

Citation metadata

Date: Jan. 7, 2014
From: PLoS ONE(Vol. 9, Issue 1)
Publisher: Public Library of Science
Document Type: Article
Length: 7,074 words
Lexile Measure: 1390L

Document controls

Main content

Article Preview :

Author(s): Bumhee Park 1,2, Dae-Shik Kim 3, Hae-Jeong Park 1,2,*


Decades of neuroimaging studies have demonstrated that cognition is co-localized with cyto and/or myeloarchitectonically distinct brain areas. Yet, more recent data suggest this structure-function relationship to be highly complex such that a single cognitive function can recruit multiple distributed local clusters of neurons [1], [2]. Furthermore, diverse brain states and functions appear to be encoded by altering connectivity among distributed neuronal clusters [3]-[5]. The importance of these distributed interactions (i.e., a network) in constructing diverse cognitions is widely acknowledged in the field of systems neuroscience.

Despite a large amount of growth in phenomenological data supporting this network perspective on cognition, the mechanism behind how the brain formulates highly diverse brain processes or characterizes various individuals has not been sufficiently studied. Given that there is a potentially infinite number of different cognitive processes, does the brain generate new networks each time it computes a new cognitive process? Recent data suggest, alternatively, that pre-existing repertoires of a much smaller number of canonical network components are selectively and dynamically recruited for various cognitions [6]. To this end, relatively well-defined network components such as working memory circuits, motor circuits, and language circuits may simply be members, or mixtures of members, of these repertoires of functional network components.

The primary aim of this study was to identify independent cognitive network components from limited sets of neuroimaging data. Instead of relying on task-specific data which are impractical to cover whole brain processes, we focused on recent findings that the pool of cognitive network components are embedded in spontaneous activity [7], independent of specific cognitive tasks, in the fashion of slow fluctuations in synchrony of distributed regions during the resting state [8].

To identify intrinsic cognitive network components from spontaneous activity, we proposed a subnetwork decomposition method from multitudes of whole brain networks, with an assumption that repertoires of intrinsic subnetworks constitute individual brain networks with different strength combinations (Figure 1). We identified intrinsic functional subnetworks of the human brain by applying independent component analysis (ICA) [9] to a group of brain networks in the form of graphs (graph-ICA) (Figure 2). Using derived subnetwork repertoires, we decomposed brain networks during specific tasks including motor activity, working memory exercises, and verb generation, and identified subnetworks associated with performance on these tasks. We also analyzed sex differences in utilization of subnetworks, which was useful in characterizing group networks.

Figure 1. Motivation for use of graph-ICA. The graph-ICA is to decompose intrinsic subnetworks based on the neurocognitive network model with two assumptions; 1) A single edge (i.e., functional connection between two regions) can be engaged in multiple cognitive functions and can be part of multiple functional subnetworks with different weights (i.e., connectivity) (B ), rather than a part of only a single subnetwork (A ); 2) Whole brain networks (i.e., graphs) can be composed of independent canonical subnetworks. Each individual recruits different subnetworks with different strengths of their involvements (C ). The usage strengths of subnetworks can be used to...

Source Citation

Source Citation   

Gale Document Number: GALE|A478874649