The efficacy of multisite MRI scanners for total brain volume measurements: a cross-sectional study in Saudi Arabia

Accurate estimation of brain structure and volume is essential in multi-scanner MRI studies investigating neurological condition. This study examines the effect of cross-scanner differences, incorporating field strength, vendor, and sequence variations, on brain volumetric measurements in Saudi adults.

to evaluate the agreement, reliability and the magnitude and direction of the systematic bias of volumetric measurements between two MRI systems (1.5 T and 3 T) differing in field strength, vendor, and sequence using a within-subject same-day design.

38 healthy participants (25.8 ± 5.6 years; 52.6% men) were scanned back-to-back on 1.5 T GE OPTIMA 450W (SPGR) at 3 T Siemens Skyra (MPRAGE). The images were analyzed using FreeSurfer v7.1.1 (longitudinal stream). To evaluate mean-level bias, a series of paired t-tests, percent differences, and false-discovery-rate (FDR) - adjusted p-values were computed. To evaluate order rank reliability, we used Pearson correlations, and two-way mixed-effects intraclass correlation coefficients [ICC(3,1)].

The global volume measurements were nearly the same on both scanners (total brain: 1 185 ± 99 mL at 1.5 T vs.1 179 ± 103 mL at 3 T; -0.6% mean difference, d = 0.07; FDR-p = 0.93) with excellent reliability (ICC = 0.993). The volumes of cortical GM, total GM, supratentorial, and cerebral WM showed a bias < 1.8% with excellent ICCs ≥ 0.978. On the contrary, some subcortical structures had a strong effect of cross-scanner differences after FDR correction: putamen (+13.1%, d = 1.26), amygdala (+10.1%, d = 0.95), accumbens (+36.1, d = 1.58), and thalamus (-6.6%, d = 0.70). Nevertheless, reliability of these nuclei remained good to excellent (ICC = 0.82-0.95). Some of the corpus-callosum subregions showed mixed directional changes (central: -22.0%; anterior: +8.6%) but had good reliability (ICC = 0.76-0.93).

There is high reliability and agreement for whole-brain, cortical, and most subcortical measures between the two MRI scanners under controlled acquisition and standardized processing conditions. Systematic, region-specific biases, especially in the putamen, amygdala, accumbens, and thalamus, necessitate harmonization prior to pooling multisite data.