Paired 3 T and 7 T MP2RAGE Brain Volumetry: Preserved Rank Order, Region-Dependent Offsets, and Exploratory Cross-Field Calibration

Structural MRI volumes from different field strengths may correlate across individuals yet differ systematically, complicating pooling and calibration. We assessed paired 3 T and 7 T MP2RAGE volumetry in 22 healthy right-handed male volunteers (of 32 initially processed; 10 excluded after FreeSurfer surface-reconstruction failure), using SynthStrip and SynthSeg on reconstructed UNI images. Volumes were normalised to total intracranial volume and evaluated for global tissue classes, subcortical structures, and mean hemispheric cortical composites. Analyses included paired Wilcoxon tests with false-discovery-rate correction, Spearman correlations, absolute-agreement ICC(2,1), Bland–Altman analysis, and an exploratory leave-one-out comparison of three linear calibration models evaluated with symmetric mean absolute percentage error (sMAPE). Rank-order association was high for global/subcortical measures (ρ = 0.804–0.965, all p < 0.001), but agreement was region-dependent. Field-specific TIV normalisation yielded four significant negative offsets (−1.00% to −2.20%), whereas the common-denominator/raw-volume analysis yielded significant positive differences in all seven regions (+1.20% to +4.35%). ICC(2,1) ranged from 0.696 to 0.980. On the %TIV scale, all calibration models reduced error in the four offset regions, but none did so for the hippocampus. Rank order was preserved more consistently than absolute agreement; offsets should be interpreted as scanner-protocol differences rather than field-strength effects alone.