Abstract 14: Predicting Hemorrhagic Transformation in Ischemic Stroke - Very Low Cerebral Blood Volume versus Permeability Analysis

Abstract

Background and purpose: Hemorrhagic transformation is the major complication of reperfusion therapies and carries a high risk of disability and death. Very low cerebral blood volume (VLCBV) and increased blood brain barrier permeability, both markers of severe ischemia, have been proposed as imaging predictors of hemorrhage risk. We aimed to compare the prognostic power of these two approaches using data from the DEFUSE 2 study. Methods: Acute ischemic stroke patients had perfusion-diffusion MRI before and within 12 hr after endovascular therapy. Baseline cerebral blood volume (CBV) maps were generated and the volume of VLCBV (CBV<2.5 th percentile of the normal hemisphere) was calculated. Permeability was assessed using two recirculation parameters: "relative recirculation" (rR): the difference in area encompassed by the observed tissue-concentration curve and a theoretical fit of the first pass bolus; and "percent recovery" (%Recovery): the difference in signal intensity between peak bolus and average post-bolus. Parenchymal hematoma (PH) was defined using ECASS criteria. Reperfusion was defined as >50% reduction in Tmax>6sec lesion volume between baseline and post-procedure MRI. Logistic regression models were compared using Bayesian Information Criterion (BIC). Results: In DEFUSE 2, MRI prior to catheter angiography was performed in 110 patients, 59 had tPA pre-treatment and 25 developed PH. In 103 patients with technically adequate acute perfusion MR, preliminary receiver operating characteristic analysis identified >40%rR and <50%Rec as the optimal thresholds to assess the volume of tissue with increased permeability. In logistic regression, PH was associated with increased VLCBV (p=0.02) and rR permeability (p=0.05) but not %Recovery (p=0.17). The VLCBV model had better fit than rR permeability (BIC difference +2.1) and there was no significant interaction between parameters (p=0.33). Reperfusion was strongly associated with PH (p=0.01) and improved fit of the VLCBV model (BIC +7.2). Conclusions: VLCBV was a stronger predictor of early PH after reperfusion than permeability parameters in patients treated with endovascular therapy.