Frontiers in Neuroscience · Oct 16, 2026
Agreement and longitudinal repeatability of established brain volumetry pipelines across healthy controls, multiple sclerosis, and Alzheimer's disease
By Šišić N, Baraković M, Edde M et al.
Introduction Brain volume estimates are increasingly used as longitudinal endpoints in neuroimaging studies and clinical trials, but outputs from different segmentation pipelines may not be interchangeable. In this study, we performed a version-specific comparison of six established brain segmentation pipelines across healthy controls (HC; n = 20), people with multiple sclerosis (MS; n = 19), and people with Alzheimer's disease (AD; n = 50). Methods FreeSurfer v6, FreeSurfer v7.2, SAMSEG, FastSurfer v1, SIENAX/SIENA-XL, and CAT12 were evaluated. The primary aims were to quantify systematic offsets and inter-pipeline agreement in normalized baseline volume estimates, assess within-pipeline repeatability and longitudinal change-from-baseline concordance, and determine whether longitudinal processing reduced variability relative to cross-sectional processing. Gray matter, white matter, hippocampus, and thalamus were selected and normalized to intracranial volume. Agreement was assessed using intraclass correlation coefficients, while longitudinal change was evaluated using mixed-effects models and repeatability metrics. Results Baseline volume estimates differed significantly across pipelines in 11 of 12 cohort-by-tissue comparisons (FDR-adjusted p < 0.001), with the largest pipeline-dependent offsets observed for grey matter, hippocampus, and thalamus. Inter-pipeline agreement was generally strong for white matter and lower for gray matter and subcortical structures (hippocampus and thalamus), particularly in pathological cohorts. Change-from-baseline models showed no significant Pipeline × Timepoint interactions after FDR correction, indicating broadly parallel longitudinal trajectories across pipelines after baseline normalization. However, significant pipeline main effects persisted in selected tissues, especially in AD and subcortical regions, indicating residual differences in the magnitude of estimated change. Longitudinal processing reduced within-subject variability in selected settings, particularly for FreeSurfer-derived hippocampal and thalamic estimates, but did not uniformly improve stability across all pipelines or tissues. Discussion Together, these findings show that systematic offset, absolute agreement, longitudinal change magnitude, and short-interval repeatability represent distinct properties of volumetric pipelines and vary by tissue, cohort, and processing mode. Longitudinal processing reduced variability but not uniformly, highlighting pipeline- and tissue-dependent differences relevant to the interpretation of longitudinal volumetric measurements.