Mapping the Metabolic Penumbra From Standard CT Perfusion: CTP-Derived CMRO 2 /OEF Estimates Versus CBF/Tmax

BACKGROUND: CT perfusion (CTP) hemodynamic thresholds may overestimate the ischemic core (ghost core) in early presenters. We hypothesized that a metabolic model, defining the core by decreased cerebral metabolic rate of oxygen and penumbra by increased oxygen extraction fraction, estimated from standard CTP, would better predict tissue fate and early neurological outcome. METHODS: This retrospective single-center study (October 2015 to October 2025) included consecutive thrombectomy-treated patients with anterior-circulation large-vessel occlusion, admission CTP, and 48±12-hour diffusion-weighted imaging. Predefined relative cerebral metabolic rate of oxygen and oxygen extraction fraction thresholds (metabolic model) were compared with relative cerebral blood flow ≤30% and time-to-maximum (Tmax) ≥6 seconds (hemodynamic model) against diffusion-weighted imaging. Each imaging model (core and penumbra volumes) was added to clinical covariates predicting 48-hour lesion volume (linear regression) and 24-hour and discharge National Institutes of Health Stroke Scale (ANCOVA), tested by nested F -tests; imaging models were compared by paired bootstrap of cross-validated R 2 . RESULTS: Of 3011 thrombectomy patients, 227 were included (median age, 70 years; 46% male; median onset-to-CTP, 108 minutes; 162 recanalized). Relative cerebral metabolic rate of oxygen ≤30% reduced core overestimation versus relative cerebral blood flow ≤30% (bias, +2.6 versus +20.0 mL; dice, 0.20 versus 0.18; both P <0.001; reproduced in temporal validation), but >10-mL underestimation was more frequent (17% versus 6%). Relative cerebral blood flow error decreased with onset-to-recanalization delay (ρ=−0.30), unlike relative cerebral metabolic rate of oxygen (ρ=0.01). In 65 nonrecanalized patients, relative oxygen extraction fraction ≥125% outperformed time-to-maximum on bias (+14.3 versus +19.3 mL; P <0.001) and dice ( P =0.002), but not absolute error ( P =0.24). The metabolic model better explained 48-hour lesion volume (cross-validated R 2 , 0.32 versus 0.22; P =0.002). Only the metabolic model improved 24-hour ( P =0.004) and discharge ( P =0.009) National Institutes of Health Stroke Scale score prediction over clinical covariates, outperforming the hemodynamic model at discharge ( P =0.04) but not at 24 hours ( P =0.21). CONCLUSIONS: A CTP-derived metabolic model reduced ghost-core misclassification, particularly in early presenters, and better explained 48-hour lesion volume than hemodynamic thresholds, with modest added value for early neurological outcome. Positron emission tomography and external validation are warranted.