Abstract WMP61: Simultaneous Cerebral Blood Flow And Blood Volume Thresholding Produces Consistent Volumes Between Computed Tomography Perfusion Deconvolution Algorithms

Abstract

Introduction: Computed tomography perfusion (CTP) allows estimation of infarct volume in patients with acute ischemic stroke, but significant differences have been reported between CTP deconvolution algorithms. Hypothesis: We hypothesized that cerebral blood flow (CBF) and blood volume (CBV) dual thresholding estimates infarct volumes accurately and remain consistent between deconvolution algorithms. Methods: Data were from the open-source ISLES dataset, which comprised of patients with acute ischemic stroke who received admission CTP and infarct delineated from diffusion-weighted imaging (DWI) acquired <3 hours of CTP and prior to treatment. CBF and CBV maps were calculated with a model-independent (filtered Fourier Transform) and a model-dependent (box-tail model) deconvolution algorithm using identical arterial input functions and Gaussian-filtered source images. Infarct was estimated using standard CBF<30% relative to the contralateral hemisphere and a CBF<30%+CBV<38% dual threshold. Volume agreement between DWI and CTP lesions was determined by the mean difference ± standard deviation (DWI minus CTP). Consistency of CTP lesion volumes between deconvolution algorithms was characterized by the Pearson correlation (r) and a paired t-test. Results: Of 63 included patients, median DWI lesion volume was 28.0 (interquartile range: 13.3 to 57.2) ml. With model-independent and model-dependent deconvolution, respectively, mean differences were 35.7±40.4 ml and 10.9±47.6 ml using CBF<30%, and 37.1±40.2 ml and 28.5±28.6 ml using CBF+CBV thresholding. Between model-independent and model-dependent CTP lesion volumes, Pearson correlation was moderate for CBF<30% (r=0.645, p<0.001) and excellent for CBF+CBV thresholding (r=0.903, p<0.001). Paired t-tests were significant for both CBF<30% and CBF+CBV thresholding (p<0.001). Conclusions: CBF+CBV lesion volumes had improved consistency between deconvolution algorithms while achieving similar agreement to DWI lesion volumes compared to the standard CBF<30% threshold. Further adjustment of CBF+CBV thresholds are required to improve accuracy to DWI lesion volumes but show promise in standardizing CTP lesion volumes across deconvolution algorithms.