Comparing longitudinal processing pipelines for brain‑volume consistency in T1‑weighted MRI test‑retest scans

Neurodegenerative diseases require longitudinal assessment to track disease progression, with brain volume change from T1-weighted MRI serving as a key biomarker that demands robust and precise processing methods. Although several longitudinal processing pipelines exist, there is no consensus on which offers the highest reliability. In this study, we evaluate six widely used open-source tools for cross-sectional and longitudinal processing of T1-weighted MRI: FreeSurfer, SAMSEG, ANTs, ANTsPyNet, SPM12, and CAT12. We assess their robustness using test-retest data from the MIRIAD cohort, in which no meaningful anatomical change is expected between repeated scans. Our results show that, overall, longitudinal processing methods demonstrate greater robustness than their cross-sectional counterparts. However, this pattern is not consistent across all tools: some longitudinal implementations do not outperform their cross-sectional versions, and the magnitude of improvement varies by method and brain region. We conclude that while the existing longitudinal processing approaches can improve consistency in brain volume estimation, these benefits are method-dependent.