Arterial Spin Labeling Reveals Disrupted Brain Networks and Functional Connectivity in Drug-Resistant Temporal Epilepsy

Abstract

Resting-state networks (RSNs) and functional connectivity (FC) have been increasingly exploited for mapping brain activity and identifying abnormalities in pathologies, including epilepsy. The majority of studies currently available are based on blood-oxygenation-level-dependent (BOLD) contrast in combination with either independent component analysis (ICA) or pairwise region of interest (ROI) correlations. Despite its success, this approach has several shortcomings as BOLD is only an indirect and non-quantitative measure of brain activity. Conversely, promising results have recently been achieved by arterial spin labeling (ASL) MRI, primarily developed to quantify brain perfusion. However, the wide application of ASL-based FC has been hampered by its complexity and relatively low robustness to noise, leaving several aspects of this approach still largely unexplored. In this study, we firstly aimed at evaluating the effect of noise reduction on spatio-temporal ASL analyses and quantifying the impact of two ad-hoc processing pipelines (basic and advanced) on connectivity measures. Once the optimal strategy had been defined, we investigated the applicability of ASL for connectivity mapping in patients with drug-resistant temporal epilepsy vs. controls (10 per group), aiming at revealing between-group voxel-wise differences in each RSN and ROI-wise FC changes. We first found ASL was able to identify the main network (DMN) along with all the others generally detected with BOLD but never previously reported from ASL. For all RSNs, ICA-based denoising (advanced pipeline) allowed to increase their similarity with the corresponding BOLD template. ASL-based RSNs were visibly consistent with literature findings; however, group differences could be identified in the structure of some networks. Indeed, statistics revealed areas of significant FC decrease in patients within different RSNs, such as DMN and cerebellum (CER), while significant increases were found in some cases, such as the visual networks. Finally, the ROI-based analyses identified several inter-hemispheric dysfunctional links (controls > patients) mainly between areas belonging to the DMN, right-left thalamus and right-left temporal lobe. Conversely, fewer connections, predominantly intra-hemispheric, showed the opposite pattern (controls < patients). All these elements provide novel insights into the pathological modulations characterizing a "network disease" as epilepsy, shading light on the importance of perfusion-based approaches for identifying the disrupted areas and communications between brain regions.

Keywords: ICA; arterial spin labeling; epilepsy; functional connectivity; perfusion; resting-state.

FIGURE 1

(A) Distribution of temporal signal-to-noise ratio (tSNR) values derived from control’s data, with both basic and advanced cleaning pipelines. The four ASL runs were considered separately (^∗∗p < 0.01, Bonferroni-corrected for multiple comparisons). (B) Loss of temporal degrees of freedom (tDoF) for each of the four runs. Only the advanced pipeline was considered, as no lost tDoFs were present in the case of the basic cleaning. (C) Spatial pattern of changes in ASL signal standard deviation after using the advanced pipeline. The probability maps, representing areas where the variance was reduced more frequently in controls, are reported for each run separately. These maps represent for each voxel the percentage of subjects with %ΔSTD > 25%.

FIGURE 2

Example of three resting-state networks (DMN, SMN, and FP.l) derived from separate healthy subjects by applying the two different cleaning pipelines to the ASL datasets. The corresponding BOLD template from literature is also reported for reference. All the component maps were thresholded at z = 3.

FIGURE 3

Distribution of Dice similarity coefficient and spatial cross-correlation values across healthy controls. The two indices were calculated for each spatial component derived from the data cleaned with the basic and advanced pipelines, respectively, by comparing the network of interest to the corresponding template component. ^∗Statistically significant differences, p < 0.05.

FIGURE 4

Mean cerebral blood flow (CBF) maps in physiological units (ml/100 g/min) calculated across the group of healthy controls from the ASL data cleaned with the basic and advanced pipelines. Some representative slices of interest in the 2-mm MNI152 standard space are reported in radiological convention. No statistically significant changes were detected between the CBF maps derived from the two pipelines when compared on a voxel-wise basis.

FIGURE 5

Twelve resting-state networks (RSNs) of interest derived from the control and patient group analyses. All spatial maps were converted to z-statistic images and thresholded at z = 3. Each map was superimposed to the 2-mm MNI152 standard space and shown in radiological convention.

FIGURE 6

Within-network functional connectivity (FC) differences between patients and controls resulting from the permutation testing on the dual regression outputs. Clusters of significant difference (p < 0.05, FWE-corrected with TFCE) are overlaid on the MNI152 standard space and shown in radiological convention. In particular, areas featuring decreased FC in patients are reported with the blue-light blue colormap (HC > PT), while increased FC patterns in patients are decoded with the red-yellow colormap (HC < PT).

FIGURE 7

Functional connectivity (FC) matrices averaged across all subjects for the two groups (controls [HC] and patients [PT]). Thirty regions of interest were extracted from the 12 resting-state networks, and were sorted in the matrices according to their belonging network. Regions are the row and column indices of the FC matrices, and the corresponding matrix element provides the color-coded Pearson correlation coefficient. The significant FC links (p < 0.05, both without correction for multiple comparisons and FWE-corrected) are also reported for the control vs. patient analysis. Each element of these matrices provides the color-coded p-value of the statistical differences, as indicated by the colorbar (blue: HC > PT; red: HC < PT). Only the lower triangular part of the matrix, i.e., excluding self-connections and redundant connections, is shown due to the symmetry of the FC matrix. MedV, medial visual; LatV, lateral visual; OccV, occipital visual; mPFC, medial prefrontal cortex; pIPL, posterior inferior parietal lobule; PCC, posterior cingulate cortex; MTG, middle temporal gyrus; Hipp, hippocampus; Cer, Cerebellum; PRG, precentral gyrus; SMA, supplementary motor area; STG, superior temporal gyrus; ACC, anterior cingulate cortex; lPFC, lateral prefrontal cortex; PPC, posterior parietal cortex; TPL, temporal lobe; THL, Thalamus; l, left; r, right.