Graph-theoretical analysis of resting-state fMRI in pediatric obsessive-compulsive disorder

Abstract

Background: fMRI graph theory reveals resting-state brain networks, but has never been used in pediatric OCD.

Methods: Whole-brain resting-state fMRI was acquired at 3T from 21 children with OCD and 20 age-matched healthy controls. BOLD connectivity was analyzed yielding global and local graph-theory metrics across 100 child-based functional nodes. We also compared local metrics between groups in frontopolar, supplementary motor, and sensorimotor cortices, regions implicated in recent neuroimaging and/or brain stimulation treatment studies in OCD.

Results: As in adults, the global metric small-worldness was significantly (P<0.05) lower in patients than controls, by 13.5% (%mean difference=100%X(OCD mean - control mean)/control mean). This suggests less efficient information transfer in patients. In addition, modularity was lower in OCD (15.1%, P<0.01), suggesting less granular - or differently organized - functional brain parcellation. Higher clustering coefficients (23.9-32.4%, P<0.05) were observed in patients in frontopolar, supplementary motor, sensorimotor, and cortices with lower betweenness centrality (-63.6%, P<0.01) at one frontopolar site. These findings are consistent with more locally intensive connectivity or less interaction with other brain regions at these sites.

Limitations: Relatively large node size; relatively small sample size, comorbidities in some patients.

Conclusions: Pediatric OCD patients demonstrate aberrant global and local resting-state network connectivity topologies compared to healthy children. Local results accord with recent views of OCD as a disorder with sensorimotor component.

Keywords: Frontal pole; Obsessive-compulsive disorder; Sensorimotor cortex; Supplementary motor cortex; fMRI graph theory.

Fig. 1

Nodes for global and local graph-theory analyses and frontopolar effects of OCD. Volume-rendered translucent brain in sagittal, axial, and coronal views displaying centers-of-gravity of 100 functional nodes for calculating global graph-theory measures. First-group nodes (see text) are green, second-group nodes red, third-group nodes blue (A). Nodes 80 (green; B) in left frontopolar cortex and 66 (yellow; C) in right frontopolar cortex had 20-35% higher group-mean clustering coefficent (indexing localized functional connectivity) in pediatric OCD patients than in healthy control children. A third node (56, not shown) in left frontopolar cortex had higher clustering coefficient and lower (-63.6%) betweenness centrality (indexing pass-through functional connectivity) in OCD than in controls. Other sites of elevated clustering coeffivcient included midline supplementary motor cortex and left sensorimotor cortex.

Fig. 2

fMRI correlation matrices and histograms. Matrices representing Pearson r for correlations between all 100 anatomically arrayed functional nodes during 8-min whole-brain BOLD rsfMRI acquisition. Values averaged across healthy control (A) and OCD (B) groups. Color bars indicate magnitude of r. Histograms of r values for control (C) and OCD (D) groups. The two group distributions were normal and overlapping.

Fig. 3

Effects of OCD and OCD symptoms on global graph-theory metrics. Group-mean values for the global rsfMRI network connectivity parameters small-worldness (A) and modularity (B) across the 10-20% of strongest connections (sparsity) for children with (filled circles) and without (open circles) OCD. Note significantly lower small-worldness (for sparsities >15%) and modularity (all sparsities) for the OCD group. Significant between-group differences at individual sparsity values are marked (*P<0.05, **P<0.01). For the overall areas-under-the-curve (AUCs), small-worldness was lower for the OCD group at P<0.05 and modularity was lower at P<0.01 (Table 2). Small-worldness findings are consistent with less efficient global information transfer in the brain during the resting state in OCD; modularity findings are consistent with the brain being divided into fewer or into a different set of functional modules during the resting state in OCD. Within the pediatric OCD group, the CY-BOCS Obsessions subscore correlated positively with small-worldness (Spearman r=0.56, P=0.008) (C).

Fig. 4

Effects of OCD and on local frontal polar graph-theory metrics. Single-subject (filled circles for patients, open circles for controls) and group-mean (bars) values for the local resting-state fMRI graph-theory metric clustering coefficient at left (Node 80) and right (Node 66) frontal poles for children with and without OCD. The OCD sample had significantly higher clustering coefficient at both sites. Higher clustering coefficient implies greater local connectivity of these nodes during the resting-state in OCD.