Resting-state function can be assessed using arterial spin labeled (ASL) perfusion MRI¹ both at the voxel level with absolute CBF and through correlations between spontaneous CBF fluctuations within functional networks. To determine the optimal conditions for resting-state ASL applications such as measuring pharmacological effects on brain function, we evaluated the reproducibility and reliability of CBF and functional connectivity (FC) metrics derived from the CBF time series across four different conditions.

Scanning protocol

Fifteen subjects (8 female, age=28 ± 5 years) consented to the study and twelve subjects completed two sessions spanning a mean of 7 days, conducted at the same time of day for each subject.

The scanning protocol, conducted on a 3T Siemens Trio with a 32-channel head-array, included a T₁-MPRAGE acquisition at 1mm isotropic resolution, a TOF angiogram to determine the optimal position of the labeling plane, and four 8-min ASL scans obtained with ‘eyes open’, ‘eyes closed’, ‘fixation’ (eyes fixated on a cross), and ‘PVT’ (psychomotor vigilance task) carried out in pseudorandomized order across subjects and sessions. A single-shot, background-suppressed pCASL sequence using an accelerated 3D RARE Stack-Of-Spirals readout² was employed to collect whole-brain perfusion maps at 3.75mm isotropic resolution. An M0 image with no ASL preparation and long TR was collected at the middle and end of the sessions.

Data processing and global CBF analysis

Raw ASL images were realigned and registered to the anatomical dataset using FSL, FreeSurfer, and custom scripts in Matlab. Subtracted pairs were converted into CBF units following the one-compartment model³. T₁-anatomical images were segmented into GM and WM tissue maps using SPM’s New Segment tool to extract mean GM, WM and global CBF values, and to compute the temporal SNR and mean GM-WM contrast ratio for each subject, session and condition. Reproducibility and reliability of mean CBF across sessions and conditions were assessed using the within-subject coefficient of variation (wsCV) and the intra-class correlation coefficient (ICC), respectively. Differences across sessions and conditions were assessed via permutations tests. Voxel-wise mean CBF differences across sessions were evaluated using a repeated-measures ANOVA with ‘session’ and ‘condition’ factors.

FC analysis

CBF time series were denoised through nuisance covariate regression (12 motion regressors, and 5 WM and CSF regressors extracted with PCA) after demeaning and detrending. FC network analysis was performed as in (⁴). Briefly, for each subject and session, five voxel-wise seed-based correlation analyses were carried out to extract: DMN (seed=PC), Attention (seed=left IPS), Motor (seed=left M1), Auditory (seed=left A1) and Visual (seed=left V1) networks. Correlation maps were transformed into z-scores using the Fisher’s transform. To assess within-network connections reliability, each network was characterized using 2-6 seeds^4,5 (r=6mm) and a correlation matrix was computed containing the z-scores of the correlations between seeds. The ICC of each of the connections represented in the matrix was then calculated across sessions, and averaged to derive a network ICC. All network ICCs were averaged to obtain a global FC ICC per condition. Group connectivity maps were calculated for each network by entering each subject’s session-averaged z-map into a one-sample t-test (p < 0.05 FWE), after registration to the MNI space.

Global CBF

No differences were found in global CBF across sessions (Fig. 1), and excellent within-session agreement was observed (wsCV<5%, ICC>0.9). Significant differences were found in both reproducibility and reliability measures across conditions (Fig. 2); Mean CBF during ‘eyes closed’ was found to be 2.3% higher than the other conditions. Higher wsCV and lower ICC were found for ‘fixation’ mean CBF when compared to the other conditions (except for ICC comparison of ‘fixation’ vs ‘eyes open’, where p=0.08), which translates into lower reproducibility and reliability compared to the other conditions. This was not accompanied by a concurrent decrease in temporal SNR or a change in GM-WM contrast ratio (Fig. 3). The repeated-measures ANOVA did not yield any significant effect of condition on regional CBF.

FC

Group connectivity maps derived for each condition are shown in Fig. 4. Slight differences in within-network mean z-scores were observed across conditions (Fig. 5A). Within-network reliability showed a decreasing trend for ‘fixation’ (Fig. 5B), in line with the results for global mean CBF, but contrary to what has been reported for resting-state BOLD⁴. Global FC ICC was significantly higher for 'eyes open' than for 'fixation', in agreement with recent work reporting greater reliability for ‘eyes open’ when compared to ‘eyes closed’ for ASL and BOLD networks⁶.

These findings suggest that ‘fixation’ is inferior to the other conditions tested for resting-state ASL reproducibility, and there was no benefit of low-level attentional task versus unconstrained rest.