Spatial confluence of psychological and anatomical network constructs in the human brain revealed by a mass meta-analysis of fMRI activation

Abstract

It is well-known that the brain's activity is organized into networks but it is unclear how many networks exist. Additionally, there is also a risk of ambiguity since different names for the same network are frequently reported in the literature. In this study, we employed a mass meta-analysis of fMRI data associated with network constructs originating from both psychology and neuroscience. Based on the results from the meta-analysis, we derived a spatial similarity map between all construct terms, showing that the brain's networks cluster hierarchically into several levels. The results presented are useful as a first step in developing a unified terminology for large-scale brain network and a platform for a queryable network atlas.

Figure 1. Clustering of spatial similarity between network constructs.

(a) Similarity matrix O_ij displaying the degree of similarity between network terms after pruning. (b) An edge-weighted spring embedded layout of (a), where each circle denotes a network term. The size of the circle is determined by the weighting terms w, which is the difference between how much network term i overlapped with other terms. Candidate BNCs are marked with a thicker black border and have their names shown. An interactive version of Figure 1b can be downloaded at:  https://github.com/wiheto/MMA_of_brain_networks (c). Average clustering coefficient of the similarity matrix (marked as ‘empirical’) versus the average clustering coefficient of the permuted null model at p = 0.001. The asterisk marks that the similarity matrix had a significant average degree of clustering compared to the null graph.

Figure 2. Hierarchical clustering of the spatial similarity of network constructs in the brain.

(a) Tree plot showing three levels of branching. Background color depicts the first level of clustering. (b) Masks of all voxels included in any of the terms within the first level of clustering. The name given for each first level cluster in Fig. 2b is the BNC with the largest weighting factor w of the first level clusters, with the exception of C7 which did not included a BNC and instead shows the largest w of any of the terms. C6 was ambiguous in both spatial patterns and collection of network terms and was left unclassified. (c) Spatial overlap of all seven first level clusters projected onto the cortex surface. Colors denote how many of the first level cluster masks that are present at any particular voxel. All colors indicate at least 2 clusters are present.

Figure 3. Binary masks of the hierarchical clustering of network constructs (see also Fig. 4).

Masks are only shown if ref. they encapsulate more than one network term, included network terms and a spatial map that was judged to be reasonable. Due to the subjective nature of ref. , the excluded maps are shown in Supplementary Figure 2. The naming of the clusters is given to the left and reflects the cluster assignment where a period indicates a new cluster level (also illustrated by the indentation in the Figure). To the right of each spatial mask is a pseudo-word cloud included of the terms that are contained within each cluster and where size and color intensity reveal which terms relative to the other terms overlap with most other terms in the cluster (weighted by w, see also Methods section). The calculation of the weighting factor, w and the scaling of w was done independently for each cluster at each level.

Figure 4. Same as Fig. 3 for the remaining clusters with branches starting at C4, C5 and C7.