Prediction of poststroke hemorrhagic transformation using computed tomography perfusion

Abstract

Background and Purpose: Intracerebral hemorrhage is a serious potential complication of stroke thrombolysis. We investigated the optimal computed tomography perfusion (CTP) parameter to predict cerebral parenchymal hematoma (PH) in acute ischemic stroke. Methods: Patients with hyperacute ischemic stroke had whole-brain CTP and follow-up computed tomography/MRI to identify hemorrhagic transformation. The association of the 3 parameters relative cerebral blood flow, relative cerebral blood volume, and time to maximum (T max) with PH was examined using receiver operating characteristic analysis and multivariate logistic regression. Results: Of 132 patients, 70 were treated with thrombolysis, and 14 (10.6%) developed PH on follow-up imaging. Baseline National Institutes of Health Stroke Scale score (P=0.033) and thrombolysis (P=0.003) were both predictive of PH. Receiver operating characteristic analysis revealed that Tmax>14 s (area under the curve=0.748; P=0.002) and relative cerebral blood flow <30% of contralateral mean (area under the curve=0.689, P=0.021) were the optimal thresholds, and the Bayesian information criterion (+2.6) indicated that Tmax was more strongly associated with PH than relative cerebral blood flow. Tmax >14 s volumes of >5 mL allowed prediction of PH with sensitivity of 79%, specificity of 68%, and negative likelihood ratio of 3.16. Tmax>14 s volume and thrombolysis were both independently predictive of PH in a multivariate logistic regression model (P<0.05). Conclusions: Tmax >14 s was the CTP parameter most strongly associated with PH. This outperformed relative cerebral blood flow <30%, which closely equates to CTP estimates of ischemic core volume. Although ischemic core volume on CTP is useful in the pretreatment prediction of PH, severe hypoperfusion on T max is more strongly associated and may allow better prediction of the likely anatomic location of hemorrhage. © 2013 American Heart Association, Inc.