Pipeline Dependence of Structural Brain MRI Morphometry: A Systematic Review of Agreement Across Software, Versions, and Processing Choices

BACKGROUND: Morphometric measurements from structural brain MRI are sensitive to analysis pipelines and software versions. PURPOSE: To evaluate inter-pipeline agreement, reliability, and systematic measurement differences across automated morphometric software packages on identical structural MRI datasets. STUDY TYPE: Systematic review (PROSPERO CRD420261456493). POPULATION: One hundred eleven studies comprising healthy and clinical cohorts across the lifespan. FIELD STRENGTH/SEQUENCE: Predominantly 1.5 T and 3 T, with 7 T data also included. ASSESSMENT: Automated and semi-automated processing pipelines (predominantly FreeSurfer, FSL, and SPM/CAT) evaluated for brain volumes and cortical thickness. STATISTICAL TESTS: Intraclass correlation coefficients (ICC), Pearson/Spearman correlation (r), Lin's concordance correlation coefficient (CCC), and Bland-Altman bias (p < 0.05). RESULTS: Agreement varied across anatomical regions, software pairs, versions, and processing conditions. Strong associations coexisted with systematic measurement differences in several comparisons, but neither high association nor high agreement was consistent across global and regional outputs. Manual-reference comparisons showed structure- and population-dependent volume overestimation, with rankings differing by outcome metric. Cortical thickness, intracranial-volume normalization, and normative percentiles showed additional method-dependent differences. Software updates and preprocessing choices altered measurements and, in some studies, downstream findings. Computational repeatability, anatomical accuracy, and longitudinal-change agreement represented distinct properties and did not consistently favor the same method. DATA CONCLUSION: High correlation or consistency does not ensure absolute agreement or interchangeability in MRI morphometry. Regional discrepancies vary across software pairs, versions, and processing conditions, limiting generalization of individual comparisons to other methods or populations. EVIDENCE LEVEL: 1. TECHNICAL EFFICACY STAGE: Stage 2.