Test-retest reliability of time-varying patterns of brain activity across single band and multiband resting-state functional magnetic resonance imaging in healthy older adults

Abstract

Leading Eigenvector Dynamics Analysis (LEiDA) is an analytic approach that characterizes brain activity recorded with functional Magnetic Resonance Imaging (fMRI) as a succession of discrete phase-locking patterns, or states, that consistently recur over time across all participants. LEiDA allows for the extraction of three state-related measures which have previously been key to gaining a better understanding of brain dynamics in both healthy and clinical populations: the probability of occurrence of a given state, its lifetime and the probability of switching from one state to another. The degree to which test-retest reliability of the LEiDA measures may be affected by increasing MRI multiband (MB) factors in comparison with single band sequences is yet to be established. In this study, 24 healthy older adults were scanned over three sessions, on weeks 0, 1, and 4. On each visit, they underwent a conventional single band resting-state fMRI (rs-fMRI) scan and three different MB rs-fMRI scans, with MB factors of 4, with and without in-plane acceleration, and 6 without in-plane acceleration. We found test-retest reliability scores to be significantly higher with MB factor 4 with and without in-plane acceleration for most cortical networks. These findings will inform the choice of acquisition parameters for future studies and clinical trials.

How do tasks impact the reliability of fMRI functional connectivity?

“…Though higher MB factors may worsen SNR (Risk et al, 2021) and alter the reliability of FC measures, prior work has assessed the effect of varying MB factors on FC reliability. The literature suggests a MB acceleration factor of four improves cortical FC reliability, while a single band sequence is preferred for subcortical regions (M. Cahart et al, 2023;M.-S. Cahart et al, 2022). Further, the combination of multiband and multiecho sequences promise to yield higher reliability estimates than MB sequences alone (Cohen et al, 2021;Fazal et al, 2023;Lynch et al, 2020) (Finn, 2021;Meer et al, 2020;Sonkusare et al, 2019), thus leading to increases in reliability.…”

“…The literature suggests a MB acceleration factor of four improves cortical FC reliability, while a single band sequence is preferred for subcortical regions (M. Cahart et al, 2023; M.‐S. Cahart et al, 2022). Further, the combination of multiband and multiecho sequences promise to yield higher reliability estimates than MB sequences alone (Cohen et al, 2021; Fazal et al, 2023; Lynch et al, 2020).…”

Reliability of variability and complexity measures for task and task‐free BOLD fMRI

“…It may nevertheless be a wise strategy to strive for acquiring as much data as possible (in terms of blocks, trials, and/or time frames) per task. Several strategies exist to increase reliability of fMRI measures that go beyond acquiring more data (Dubois & Adolphs, 2016 ; Elliott et al, 2021 ), including the task design (Bennett & Miller, 2013 ; Gorgolewski et al, 2013 ; Rai et al, 2024 ), the use of multi‐echo or multiband fMRI (Cahart et al, 2022 ; Lynch et al, 2020 ), scrubbing for noise reduction (Phạm et al, 2023 ), data modeling (Hu et al, 2023 ), or functional network construction techniques (Jiang et al, 2021 ). Our results further suggest that using global indicators of brain signal variability and complexity, averaged across brain regions, appears a viable strategy to increase reliability.…”

“…It may nevertheless be a wise strategy to strive for acquiring as much data as possible (in terms of blocks, trials, and/or time frames) per task. Several strategies exist to increase reliability of fMRI measures that go beyond acquiring more data (Dubois & Adolphs, 2016;Elliott et al, 2021), including the task design (Bennett & Miller, 2013;Gorgolewski et al, 2013;Rai et al, 2024), the use of multi-echo or multiband fMRI (Cahart et al, 2022;Lynch et al, 2020) (Krohn et al, 2023). On the other hand, some measures showed widely varying (including very low) reliability estimates across regions and functional networks for some tasks (like FE BOLD for the gambling task; Figure 2b,c, right), but still moderate to good global reliability in the respective data set (Figure 2a).…”