Automated MRI-based brain volumetry demonstrates excellent intraindividual robustness in short-term follow-up despite pronounced regional variability

Background Although automated MRI-based brain volumetry analysis (AVA) tools have been suggested for objective quantification of brain regions in patients with suspected dementia, their robustness in real-life longitudinal clinical cohorts remain uncertain. Therefore, we investigated a well-established, FDA-approved AVA tool on short-term imaging follow-ups (≤1 year), expecting minimal volumetric changes, within a real-world heterogeneous cohort to specifically quantify scanner and protocol effects on the AVA tool’s performance. Methods 44 patients were retrieved from local RIS/PACS, each undergoing two 3D T1-weighted MRI scans within a one-year period. Absolute volumetric results were normalized to the respective regional baseline (delta). Intraindividual reliability was assessed using intraclass correlation coefficients (ICCs). Mean delta differences were evaluated using paired t-tests, variance differences across brain regions and scanner groups with Levene’s test, and measurement precision with the Maloney–Rastogi test (MAL) assessing intraindividual variability over follow-up relative to population-level variance. Results Intraindividual volumetric results in 44 imaging pairs (N _female =29/44, 66.0%; mean age = 64.5 yrs, median follow-up = 0.53, LQ-UQ: 0.16–0.9, range: 0.03–1.05 years) showed excellent ICCs (median = 0.93, LQ-UQ: 0.9–0.96, range 0.85–0.99). Most parenchymal regions showed no significant differences in mean (p _T ) or standard deviation (p _MAL ) between the two examinations, except for the right occipital white matter (p _T =0.01), total caudate volume (p _T =0.04), and right (p _MAL =0.02) and left frontal gray matter (p _MAL =0.04). Deltas across all brain regions showed the widest variation when scans from external facilities were included (range − 38.4% to + 68.8%). Notably, significant differences in mean (p _m ) and variance (p _v ) in regions relevant to dementia diagnosis were observed for total gray matter and cortical GMs of both hemispheres as well as both frontal and temporal lobes. Conclusions The AVA tool provided reliable intraindividual measurements, however, regional volume changes varied substantially, particularly when different scanners were included, likely due to differing technical settings. Longitudinal imaging should, therefore, follow highly standardized protocols, ideally on the same scanner with identical protocols to ensure consistency.