The semantic anatomical network: Evidence from healthy and brain-damaged patient populations

Abstract

Semantic processing is central to cognition and is supported by widely distributed gray matter (GM) regions and white matter (WM) tracts. The exact manner in which GM regions are anatomically connected to process semantics remains unknown. We mapped the semantic anatomical network (connectome) by conducting diffusion imaging tractography in 48 healthy participants across 90 GM "nodes," and correlating the integrity of each obtained WM edge and semantic performance across 80 brain-damaged patients. Fifty-three WM edges were obtained whose lower integrity associated with semantic deficits and together with their linked GM nodes constitute a semantic WM network. Graph analyses of this network revealed three structurally segregated modules that point to distinct semantic processing components and identified network hubs and connectors that are central in the communication across the subnetworks. Together, our results provide an anatomical framework of human semantic network, advancing the understanding of the structural substrates supporting semantic processing.

Keywords: connectomics; diffusion tensor imaging; module; semantics; white-matter network.

Figure 1

A flowchart for the construction of the semantic WM network. ① Transforming the AAL atlas in the MNI space to the native diffusion space by applying the inverse transformation obtained from the native T1‐weighted image (a), resulting in a subject‐specific AAL mask in the DTI native space (c). ② Reconstructing all WM fibers (d) in the whole brain from the native FA image (b) using DTI deterministic tractography. ③ Determining the WM fibers connecting every pair of gray matter regions for each healthy subject. All tracts in the native space were transformed to the MNI space. ④ Identifying the network matrix (e, black: 1; white: 0) and building binary tract maps. The inset shows a 3D view of an exemplar tract map (f), which connects the orbital part of the inferior frontal gyrus and the middle temporal gyrus. ⑤ Calculating the lesion volume in a tract for each patient by overlapping the binary tract map with the lesion image (g) in the MNI space. ⑥ Calculating the mean FA in a tract for each patient by overlapping the binary tract map with the FA image (h) in the MNI space. ⑦ ⑧ Correlating the lesion volumes/FA values and the semantic PCA score (extracted from eight behavioral tasks), controlling for confounding variables (see Method for details) in each tract and constructing weighted lesion network matrices (i) and FA network matrices (k) using partial correlation coefficients. ⑨ ⑩Transforming the weighted network matrices into binary network matrices by applying statistical thresholds (FDR corrected qs < 0.05), resulting in 41 semantic tracts in the lesion network matrices (j) and 36 tracts in the FA network matrices (l). [Color figure can be viewed in the online issue, which is available at http://wileyonlinelibrary.com.]

Figure 2

Whole brain anatomical network and the semantic anatomical network. (a) Six hundred and eighty‐eight WM tracts were successfully tracked between the 90 AAL regions in 48 healthy adults, resulting in a whole‐brain anatomical network. (b) The WM tracts whose integrity values (lesion volume or mean FA value) significantly correlated with the semantic PCA scores across 80 patients, regressing out demographic variables and lesion volume variables (FDR corrected qs < 0.05). Most semantic tracts (24 tracts) were observed in both lesion volume and FA analyses (yellow); 17 were significant only in the lesion volume analysis (blue); 12 only in the FA analysis (green; see Supporting Information Table 3 for details). These edges and the gray matter nodes they connect constitute the semantic anatomical network. The network was visualized using in‐house BrainNet Viewer [Xia et al., 2013]. [Color figure can be viewed in the online issue, which is available at http://wileyonlinelibrary.com.]

Figure 3

Lesion overlap map of the 80 patients (the n value of each voxel denotes the number of patients with lesions on that voxel). [Color figure can be viewed in the online issue, which is available at http://wileyonlinelibrary.com.]

Figure 4

The results of the graph analysis for the semantic anatomical network. (a) Three separate modules shown on a two‐dimensional pseudostructural map and a three‐dimensional structural brain. The edges and nodes are colored according to their module membership. The radius of each node denotes its degree value. (b) The participation coefficients of the network nodes, ranked by value. The line denotes > 1 SD. [Color figure can be viewed in the online issue, which is available at http://wileyonlinelibrary.com.]