Abstract WP33: Reduced Cerebral Blood Flow on Acute Whole Brain CT Perfusion Best Predicts Hemorrhagic Transformation
Objective: Hemorrhagic transformation in ischemic stroke is a potentially life threatening complication of thrombolysis. Using perfusion MRI, very low cerebral blood volume (VLCBV) strongly predicts hemorrhagic transformation after reperfusion. CT perfusion (CTP) is currently more widely accessible than MRI and recent data have shown that CT relative cerebral blood flow (relCBF) provides a better estimate of infarct core than CBV. We aimed to determine the optimal parameter to predict hemorrhagic transformation using whole brain CTP.
Methods: Patients with ischemic stroke were imaged with whole brain CTP within 6hrs of symptom onset. Hemorrhagic transformation was assessed on CT/MRI within 7 days of stroke using ECASS grade. CBF and CBV were analyzed within a relative time to peak >4sec region of interest. Results were expressed as volumes below a given percentile relative to the contralateral hemisphere (relCBF and relCBV). Receiver operating characteristic (ROC) and logistic regression analysis were performed to determine the optimal parameter and percentile threshold correlating with parenchymal hemorrhage (PH).
Results: 128 patients with acute CTP were analyzed, median age 76yr (IQR 66-83), median NIHSS 13 (IQR 9-16), 59% received IV thrombolysis. 11 patients had PH on follow-up. On ROC analysis, the optimal threshold for very low CBF (VLCBF) was at the <0th centile. VLCBF was significantly associated with PH in ROC analysis (AUC=0.760, p<0.01) whereas VLCBV (AUC 0.638 at <5th centile, 0.618 at <2.5th centile, 0.440 at <0th centile) was not significant. Using VLCBF, the optimal lesion volume to predict PH was >3mL with OR 12.0 (95%CI 2.4-58), sensitivity 0.82 (95%CI 0.48-0.98), specificity 0.73 (95%CI 0.64-0.80), negative predictive value 0.98 (95%CI 0.92-1.0) and positive predictive value 0.22 (95%CI 0.11-0.38). In logistic regression, PH was associated with increased VLCBF (p<0.01) but not with VLCBV (p=0.08). The Bayesian information criterion for VLCBF compared to VLCBV was +5 indicating improved model fit.
Conclusions: VLCBF appears to be more reliably associated with hemorrhagic transformation than VLCBV when CT perfusion is used. This may be due to reduced ability of VLCBV to distinguish regions of ischemia from normal white matter.
- © 2012 by American Heart Association, Inc.