Abstract WP87: Defining Ischemic Core in Acute Ischemic Stroke: A Multiparametric Bayesian-Based Model

• Published in: Stroke (1970) Vol. 50; no. Suppl_1
• Main Authors: Nael, Kambiz, Tadayon, Ehsan, Daoud, Amy, Fifi, Johanna, Tuhrim, Stanley, Wheelwright, Danielle, Chang, Helena, Mocco, J
• Format: Journal Article
• Language: English
• Published: 01.02.2019
• ISSN: 0039-2499; 1524-4628
• DOI: 10.1161/str.50.suppl_1.WP87

Abstract only Purpose: Bayesian model has shown promising results to offset noise-related variability in perfusion analysis. Using baseline CTP in patients with acute ischemic stroke (AIS), we aim to find optimal Bayesian-estimated thresholds to construct a multiparametric model that can provide the most accurate estimate of ischemic core. Methods: AIS patients with anterior circulation stroke who had baseline CTP and achieved successful recanalization (≥ IIb IIB) were included. CTP data were processed using a Bayesian probabilistic method. Five CTP parameters including delay, TTP, MTT, CBV and CBF and 5 additional “difference maps” were generated by subtracting the mean of a cube (27 voxels) on the contralateral side of each voxel (delay diff , TTP diff , CBV diff , CBF diff , MTT diff ). Brain was extracted from CTP, coregistered with the follow up MRI, on which infarction and non-infarction masks were drawn. A robust logistic regression was performed to assess the binary outcome of voxel-based analysis (infracted vs non-infarcted) by adjusting for intra-subject correlations. Subsequently we obtained the CTP-estimated ischemic core volume from our model (logit-score) and from what is used in routine clinical practice (rCBF <30% based on singular-value deconvolution= SVD) and compared with those obtained from MRI using Bland-Altman analysis. Results: Four CTP variables (threshold/AUC) remained independent predictor of infarction: TTP (28.8/0.76); CBF (22.1/0.73); delay diff (0.87/0.80); and MTT diff (1.38/0.69). Our imaging model output summarized in a logit score ( = -3.9170 + 0.0601*TTP - 0.0095*CBF + 0.4629*ATD diff + 0.0989*MTT diff ) identified infracted voxels with overall AUC of 0.88 at score of 0.109. In a total of 58 patients included, the difference of ischemic core volume between CTP-estimation and MRI (mean, 95%CI) were (-6, -13 +2) for SVD and (+1, -3 +4) for Bayesian logit model. Bland-Altman analysis showed the limits of agreement ranging from -61 to +50 for SVD-CBF and -25 to +26 for Bayesian-logit model. Conclusion: We established CTP thresholds for Bayesian model to estimate ischemic core. Our multiparametric Bayesian-based model improves variability in CTP-estimation of ischemic core in comparison to methodology used in current clinical routine.