Advantage of CT-based classification in Chinese cohorts: SMASH-U outperforms MRI for mortality prediction in primary intracerebral hemorrhage

Underlying cause is the critical determinant of outcome after primary intracerebral hemorrhage (ICH). Causal classifications based on computer tomography (CT) and magnetic resonance imaging (MRI) are practical for clinical application, but the predictive validity in determining ICH outcomes remains unexplored. We aimed to compare the prognostic performance of two MRI-based classifications against the CT-based SMASH-U classification in determining and discriminating ICH outcomes. This study comprised two cohorts of ICH patients: a prospective, single-center cohort (Cohort 1) evaluated with both CT and MRI, and a multicenter cohort (Cohort 2) using CT only. In Cohort 1, discriminative ability was assessed and compared among the three classifications. In Cohort 2, as a sensitivity analysis, the consistency and generalizability of CT-based classification were then evaluated in a different population. Cohort 1 included 843 patients (age 61 ± 13 years, 64.3% male) with a median follow-up of 3.0 years (IQR 1.0–4.2). CT-based SMASH-U showed the least percentage of undetermined cases (20.2%) and the best discriminative ability for baseline hematoma volume ( F value: 95.06, National Institute of Health stroke scale (NIHSS) ( F value: 3.03) and long-term survival ( χ 2 = 43.3). While MRI-based classifications performed better for ICH recurrence. Cohort 2 ( N = 1570, age 66 ± 14 years, 65.4% male), with 16.43% undetermined category using CT, confirmed that SMASH-U had good discriminative ability for all-cause mortality but was inferior for ICH recurrence. These findings highlight CT’s cost-effectiveness in establishing definitive ICH etiologies and predicting mortality, while MRI provides added value for recurrence risk assessment. Clinically, prioritizing CT for initial evaluation balances practicality and prognostic utility, with MRI reserved for cases where recurrence stratification is critical. This dual approach optimizes resource allocation and outcome prediction tailored to clinical priorities. ClinicalTrials NCT06548737 (retrospectively registered).