Ultra-high-field MRI reveals a venous-dominant microbleed phenotype in cerebral amyloid angiopathy: a comparative study versus 3 Tesla MRI

BACKGROUND AND PURPOSE: Cerebral microbleeds (CMB) are core MRI markers of cerebral amyloid angiopathy (CAA). Ultra-high-field 7T MRI may improve CMB detection and visualization of venous-associated CMB (vCMB), but the extent to which this reflects additional lesion detection versus field-dependent phenotypic reclassification remains uncertain. We compared CMB burden and venous phenotyping at 7T versus 3T in patients with probable CAA. MATERIALS AND METHODS: In this single-center retrospective study, 22 patients with probable CAA underwent paired 3T and 7T MRI within 6 months. Two blinded neuroradiologists assessed total CMB, vCMB, simple/non-venous CMB (sCMB), and the vCMB-to-total ratio on susceptibility-weighted imaging. Paired comparisons used Wilcoxon signed-rank tests; inter-scanner analyses used Spearman correlations and Bland-Altman analysis. Lesion-level distributions were descriptive only because of within-subject clustering. RESULTS: 7T detected more total CMB and vCMB than 3T (1551 versus 818 and 1173 versus 385), whereas sCMB counts did not differ significantly (p = 0.47). The median vCMB-to-total ratio was higher at 7T than at 3T (0.70 versus 0.33; p < 0.001). Inter-scanner correlations were strong for total CMB (r = 0.95) and vCMB count (r = 0.88), but weak and non-significant for the vCMB-to-total ratio (r = 0.32, p = 0.16; n = 21). Inter-rater reliability for the vCMB-to-total ratio was higher at 7T than at 3T (ICC = 0.76 versus 0.54). Cohort-level delta analyses showed that the increase in vCMB approximated the increase in total CMB, whereas the change in sCMB was close to zero. CONCLUSION: 7T MRI robustly improves CMB detection in CAA and provides higher inter-rater reliability for vCMB classification than 3T. However, weak cross-scanner correlation and agreement for the vCMB-to-total ratio preclude interchangeability across field strengths. The higher 7T vCMB burden is most consistent with additional small-lesion detection, with partial reclassification also possible. 3T remains suitable for global CMB burden assessment, whereas 7T adds value for mechanistic and longitudinal studies.