Reliability of quantitative multiparameter maps is high for MT and PD but attenuated for R1 and R2* in healthy young adults

Abstract

We investigate the reliability of individual differences of four quantities measured by magnetic resonance imaging based multiparameter mapping (MPM): magnetization transfer (MT), proton density (PD), longitudinal relaxation rate (R1), and effective transverse relaxation rate (R2*). A total of four MPM datasets, two on each of two consecutive days, were acquired in healthy young adults. On Day 1, no repositioning occurred; on Day 2, participants were repositioned between MPM datasets. Using intra-class correlation effect decomposition (ICED), we assessed the contributions of session-specific, day-specific, and residual sources of measurement error. For whole-brain gray and white matter, all four MPM parameters showed high reproducibility and high reliability, as indexed by the coefficient of variation (CoV) and the intra-class correlation (ICC). However, MT, PD, R1, and R2* differed markedly in the extent to which reliability varied across brain regions. MT and PD showed high reliability in almost all regions. In contrast, R1 and R2* showed low reliability in some regions outside the basal ganglia, such that the sum of the measurement error estimates in our structural equation model was higher than estimates of between-person differences. In sum, in this sample of healthy young adults, the four MPM parameters showed very little variability over four measurements over two days but differed in how well they could assess between-person differences. We conclude that R1 and R2* might carry only limited person-specific information in samples of healthy young adults, and, by implication, might be of restricted utility for studying associations to between-person differences in behavior.

Contributions of the Catechol-O-Methyltransferase Val158Met Polymorphism to Changes in Brain Iron Across Adulthood and Their Relationships to Working Memory

“…Presence of diamagnetic elements causes R2* to overestimate iron content and QSM to underestimate it ( Haacke et al, 2005 ). Although both QSM and R2* were available, for the present study we favoured QSM over R2*, as Wenger et al (2021) recently showed poor reliability of R2* in regions outside basal ganglia (see section “MRI Acquisition and Preprocessing” Methods for further detail).…”

“…Relaxometry was also available in the study (see Salami et al, 2021 ). However, R2* has been observed to have poor reliability in regions outside the basal ganglia ( Wenger et al, 2021 ). For comparative purpose, post hoc analyses showed a good intraclass correlation (ICC; Koo and Li, 2016 for interpretation of ICC) of 0.798 for QSM in DLPFC, which was greater than R2* (0.730).…”

Longitudinal stability of cortical grey matter measures varies across brain regions, imaging metrics, and testing sites in the ABCD study

“…Moreover, impactful events that occur in between scans such as learning new skills (e.g. Wenger et al, 2021), or adverse events such as brain injury (e.g. Lindberg et al, 2019), will lead to lasting differences.…”

“…To address our first question (what is the longitudinal stability of grey matter measures in the ABCD study, and do they differ across brain regions, structural metrics, and testing sites?) we ran a series of ICED models (Brandmaier, Wenger, et al, 2018; for other applied studies, see Anand et al, 2022;Wenger et al, 2021). We estimated between-subject and error variances for three grey matter measures (cortical thickness, surface area, and volume) across regions of interest.…”

“…More commonly than not, we do not know how reliable our measures are (Flake et al, 2017;Gawronski et al, 2011;Hussey & Hughes, 2018;Parsons et al, 2019). This basic psychometric concern does not only relate to questionnaires, but also cognitive measurements (Parsons et al, 2019) and neuroimaging metrics (Anand et al, 2022;Brandmaier, Wenger, et al, 2018;Noble et al, 2017;Wenger et al, 2021;Zuo et al, 2019). Low reliability translates to low statistical power and related challenges, including a decreased likelihood that a significant finding reflects a true effect (Button et al, 2013), is in the correct direction (Type 2 "Sign" error; Gelman & Carlin, 2014), and an inherent overestimation of the true effect size (Type M "Magnitude" error; Gelman & Carlin, 2014).…”