BACKGROUND AND OBJECTIVES: Cerebral amyloid angiopathy (CAA) is a leading cause of lobar intracerebral hemorrhage (ICH) in older individuals, associated with significant morbidity and recurrence. The updated Boston criteria version 2.0 (v2.0) incorporate new MRI biomarkers to improve diagnostic accuracy. This study aimed to validate the diagnostic performance of v2.0 compared with version 1.5 (v1.5) in patients with spontaneous ICH undergoing surgical evacuation and brain biopsy. METHODS: This retrospective single-center cohort study was conducted at the Mount Sinai Health System from 2015 to 2021. Patients with spontaneous ICH who underwent surgical evacuation with brain biopsy and preoperative MRI were included. MRI markers assessed included lobar hemorrhagic lesions (ICH, cerebral microbleeds [CMBs], cortical siderosis [cSS]) and nonhemorrhagic markers (severe visible perivascular spaces in the centrum semiovale [CSO-PVS] and multispot white matter hyperintensities [WMHs]). Pathologic confirmation of CAA was based on modified Vonsattel grading, which evaluates β-amyloid deposition in vessel walls. Diagnostic performance of v2.0 was compared with v1.5 using sensitivity, specificity, and predictive values. Logistic regression models calculated odds ratios (ORs) and 95% CIs for associations between MRI biomarkers and pathologically confirmed CAA. RESULTS: Among 186 patients (median age: 63 years
38% female), 24% had confirmed CAA. The Boston criteria v2.0 demonstrated higher sensitivity for probable CAA (0.75 vs 0.57) while maintaining specificity (0.96 vs 0.99). For possible CAA, sensitivity improved modestly (0.82 vs 0.77) with comparable specificity (0.84 vs 0.87). Among hemorrhagic markers, cSS (OR 4.14, 95% CI 1.35-13.00, DISCUSSION: The Boston criteria v2.0 enhance sensitivity for diagnosing probable CAA without compromising specificity, largely due to the inclusion of nonhemorrhagic markers such as CSO-PVS. Limitations include the retrospective design, the absence of formal inter-rater reliability measures, and the modest sample size. These findings underscore the potential of v2.0 to improve the diagnostic framework for CAA.