Abstract W P49: Optimizing Ct Perfusion for Prediction of Ischemic Core in Acute Stroke With an Open-source Benchmarking Tool to Standardize Stroke Imaging Research
Background: Multiple studies have investigated the optimal CT perfusion (CTP) parameter for identification of the ischemic core. Differences in methodology have led to results that are not easily comparable. The aim of this project was to create and test a benchmarking tool that will standardize the evaluation of CTP-software. Methods: We developed a tool consisting of: 1) an imaging library and 2) a statistical analysis algorithm. The imaging library includes pre-processed DWI and CTP scans from a large cohort of 104 acute stroke patients from 2 centers who underwent DWI within 2 hours of CTP. Pre-processing included manual delineation of the acute infarct on DWI (DWI-ROIs) and motion-correction, time-correction, and co-registration of the CTP images. The statistical analysis algorithm evaluates the performance of CTP-software by determining the sensitivity, specificity and volumetric accuracy of the CTP-software’s ischemic core (CTP-ROIs). The tool was tested with CTP-ROIs based on relative CBF (rCBF) thresholds (ranging from 0.00 to 1.00 in 0.02 increments), generated by in-house developed CTP-software (RAPID). Results: We successfully pre-processed the DWI and CTP data of 104 patients. Median time between CT and start of MR was 36 min (IQR 25-77). The tool’s statistical analysis algorithm successfully determined the performance metrics of our in-house CTP-software: Volumetric accuracy of the CTP-ROIs was optimal at an rCBF threshold of 0.36. At this threshold the median absolute volumetric difference between the CTP- and DWI-ROIs was 10 ml (4-17 IQR), sensitivity 64% and specificity 90%.Conclusion: This open-source CTP benchmarking tool provides the scientific community, for the first time, with a method to directly compare the accuracy of CTP-software packages. This may lead to major improvements in CTP-software, as researchers worldwide can use it to determine the optimal algorithm for identifying the ischemic core with CTP.
Author Disclosures: C.W. Cereda: None. S. Christensen: None. B.C.V. Campbell: None. N.K. Mishra: None. C. Levi: None. M. Mlynash: None. M. Straka: Ownership Interest; Modest; Equity interest: iSchemaView. R. Bammer: Ownership Interest; Modest; Equity interest: iSchemaView. G.W. Albers: Ownership Interest; Modest; Equity interest: iSchemaView. Consultant/Advisory Board; Modest; iSchemaView, Covidien, Lundbeck. M.W. Parson: None. M.G. Lansberg: None.
- © 2015 by American Heart Association, Inc.