hide
Free keywords:
-
Abstract:
Introduction:
Resting state fMRI (rs-fMRI) studies have identified functional connectivity patterns associated with acute undernutrition in anorexia nervosa (AN), but few have investigated recovered patients and a trait connectivity profile characteristic of the disorder remains elusive (Gaudio et al., 2016). Moreover, the majority of previous studies have employed traditional analysis procedures which fail to appreciate the complex nature of brain network organization. One previous rs-fMRI study in acute AN employing graph-theoretic metrics revealed changes in global and intermediate brain network architecture possibly driven by local degradations in a thalamo-insular network (Geisler et al., 2016). To disentangle trait from starvation effects, the present study examines network organization in recovered patients.
Methods:
The study includes fMRI resting state data of 55 patients recovered from AN (recAN) and 55 age-matched healthy females (HC). FMRI data were acquired with a 3T MRI scanner (TR=2200ms, t=6:58 min) and then preprocessed within the nipype framework using SPM8 including the artifact detection (ART), DARTEL toolbox for generating a group template and spatial normalization. Then by using the DPARSFA toolbox we parcellated the volumes into 160 spherical regions of interest (ROIs) as defined by Dosenbach (Dosenbach et al., 2010). The extracted time courses of these ROIs were used to create symmetric correlation matrices with pair-wise Pearson correlation coefficients. Based on these matrices, we constructed weighted, undirected graph networks with 160 nodes on individual subject level. We computed well-established global metrics (clustering coefficient, characteristic pathlength, small-worldness index, efficiency, assortativity) and local graph metrics (degree, strength, average pathlength, betweenness centrality, participation index, local efficiency, normalized local efficiency]) across a range of network densities (Bullmore and Sporns, 2009; Hagberg et al., 2008). All metrics were subjected to between-group comparisons using non-parametric independent tests. To discriminate between recAN and HC, a linear support vector classifier (SVC) was applied to all local metrics of all 160 regions (Pedregosa et al., 2011).
Results:
Indicative of an altered global network structure, recAN showed increased assortativity and reduced global clustering as well as small-worldness compared to HC, while no group differences at an intermediate or local network level were evident. However, using support-vector classifier on local metrics, recAN and HC could be separated with an accuracy of 70.4%. Classification appeared to be driven by differences in several local graph metrics in temporo-parietal brain regions.
Conclusions:
The current study, which is the largest rs-fMRI study in recAN to date, provides evidence for an altered global brain network configuration characterized by an affinity for nodes of a similar degree to connect preferably, similar to patients at the very beginning of therapy (Geisler et al., 2016). Moreover, the network topology of recAN was characterized by a reduced presence of highly interconnected groups of nodes. While the former finding may represent a trait marker of recAN, the latter could be seen as a scar following prolonged periods of undernutrition.
