Zitat
L. Butry, J. Thomä, J. Forsting, E. Enax-Krumova, and L. Schlaffke, “Cross-vendor reliability of functional and structural brain connectivity in a travelling cohort,” Scientific Reports, vol. 16, no. 1, p. 12071, 2026.
Abstract
Scanner-related bias is a major challenge in multi-centre neuroimaging studies. This study assesses cross-vendor reliability of functional and structural brain connectivity (FC, SC) and evaluates the effect of neuroComBat harmonisation. A travelling cohort (n = 10) of healthy participants was scanned on a Siemens Prisma 3T and a Philips Achieva 3.0T system within one week. Reliability metrics of FC (Pearson’s r) and SC (based on iFOD2 & SIFT2) were calculated for each edge before and after neuroComBat harmonisation. At the subject-level, reliability was poor for FC (ICC: 0.22 ± 0.23, wsCV: 62 ± 14%, unharmonised) and fair for SC (ICC: 0.43 ± 0.32, wsCV: 28 ± 15%, unharmonised). At the group-level, Bland-Altman analyses showed minimal systematic bias between scanners in both modalities, but with wide limits of agreement. Pattern similarity was 0.63 (FC) and 0.98 (SC). G Theory analyses showed the scanner accounted for 11.8% (FC) and 7.9% (SC) of the variance. After harmonisation, subject-level reliability only improved marginally, but scanner-dependent variability dropped to 0.8% and 1.1% in FC and SC, respectively. Overall, reliability was higher for SC than FC mirroring their differential temporal stability, for group-averaged than subject-level data, and for fullconnectome organisation than for individual edges. NeuroComBat worked as designed, successfully mitigating scanner-dependent variance at the group but not the subject level.