Accelerated MRI Sequences for Intracranial Hemorrhage Screening

OBJECTIVES: Anti-amyloid-beta immunotherapy requires frequent MRI screening for amyloid-related imaging abnormalities-hemorrhage subtype (ARIA-H), consisting of cerebral microbleeds (CMB) and/or superficial siderosis (SS), using gradient-recalled echo (GRE) or susceptibility-weighted imaging (SWI). Screening MRI sequences for ARIA-H may benefit from acceleration to maximize patient enrollment by increased throughput and reduced motion degradation. This study assessed the diagnostic performance of standard GRE and SWI to echo-planar imaging (EPI) accelerated substitutions for detecting CMB and SS. MATERIALS AND METHODS: This retrospective single-center rater study included 50 patients, 25 with CMB and 25 patients without CMB (median age 77 y, IQR: 70 to 82 y; 30 of 50 female) who were imaged with FDG PET-3T MRI from April to July 2023. Standard GRE (90 s) and SWI (192 s) were compared with an EPI-accelerated GRE (aGRE; 13 s, 86% time reduction) and an EPI-accelerated SWI substitution (aSWI; 33 s, 83% time reduction). Three board-certified neuroradiologists independently reported CMB and SS (per ARIA-H monitoring guidelines), perceived image quality and motion for each sequence. There were 240 total assessments per rater (the 4 different sequences for the 50 patients plus 10 duplicated patients). Sensitivity, specificity, positive and negative predictive values, area under the curve (AUC), inter-rater and intrarater agreement were determined for each sequence and rater. RESULTS: The aggregate AUCs for the 4 individual sequences were excellent for detecting CMB (0.84 to 0.94) and SS (0.89 to 1.00) without statistical differences observed between standard and EPI-accelerated substitutions. Both aGRE and aSWI had high negative predictive values (96.5% to 100%). There were modest quantitative correlations between standard and accelerated sequences (0.606 and 0.391 for GRE and SWI, respectively), no differences in CMB count for aGRE (bias 0.01, P =0.895), but reduced CMB count with aSWI (bias -1.12, P =0.014). Inter-rater agreements were mildly reduced for both GRE versus aGRE (eg, 0.757 to 0.622 for CMB detection) and SWI versus aSWI (eg, 0.834 to 0.649 for SS detection). Perceived image quality for accelerated sequences was reduced, but with less motion observed with aSWI. CONCLUSIONS: The aGRE and aSWI sequences shorten scan times 86% and 83%, respectively, with similar diagnostic performance for ARIA-H screening, but reduced rater agreement and perceived image quality.