AJR:188, January 2007 213 AJR 2007; 188:213–218 0361–803X/07/1881–213 © American Roentgen Ray Society 213.fm — 11/30/06 Sanelli et al. CT Perfusion Maps Neuroradiology • Original Research Reproducibility of Postprocessing of Quantitative CT Perfusion Maps Pina C. Sanelli 1 Gregory Nicola 1 Apostolos J. Tsiouris 1 Igor Ougorets 2 Charles Knight 1 Bruce Frommer 1 Steve Veronelli 1 Robert D. Zimmerman 1 Sanelli PC, Nicola G, Tsiouris AJ, et al. Keywords: brain, cerebrovascular disease, CT, perfusion CT DOI:10.2214/AJR.05.2188 Received December 20, 2005; accepted after revision June 28, 2006. Supported in part by a GE-AUR Research Award. 1 Department of Radiology, New York Presbyterian Hospital, Weill Medical College of Cornell University, 520 E 70th St., Starr 630, New York, NY 10021. Address correspondence to P. C. Sanelli (pcs9001@med.cornell.edu). 2 Department of Neurology, New York Presbyterian Hospital, Weill Medical College of Cornell University, New York, NY 10021. OBJECTIVE. The purpose of this study was to assess interobserver and intraobserver vari- ability in evaluation of the reproducibility of quantitative data obtained in semiautomated post- processing of CT perfusion data sets by observers of different levels of skill and experience and in fully automated postprocessing. MATERIALS AND METHODS. Twenty CT perfusion data sets were postprocessed by a neuroradiologist using an automated postprocessing program and by five observers (neuroradiol- ogy attending, neurology attending, radiology resident, senior and junior CT technologists) who received a brief training session in use of software for semiautomated postprocessing. For assess- ment of intraobserver variability, each observer repeated postprocessing of 10 CT perfusion data sets. Standard regions of interest were placed on identical locations for each observer’s cerebral blood flow (CBF), cerebral blood volume (CBV), and mean transit time (MTT) maps of three brain regions: an ischemia–infarct region, normal cortical gray matter, and white matter. RESULTS. The variability in mean quantitative values of CBF, CBV, and MTT was 2.5–9.5% among all observers. Greater variability (20.4%) was introduced with the automated program. High correlation was found among all possible pairings of observers (r = 0.87–0.99). Low correlation was observed between automated postprocessing and postprocessing by all ob- servers. Intraobserver variability in quantitative CT perfusion data ranged from 0.29% to 10.8%. High intraobserver correlation (r = 0.91–0.99) was found for the observers. CONCLUSION. Quantitative CBF, CBV, and MTT data obtained from postprocessing of CT perfusion data sets are reproducible among observers with varying levels of skill and experi- ence. Observer interaction with the software is an important component for correct identification of user-defined parameters. Establishing a uniform and standard postprocessing technique is es- sential for maintaining good reproducibility. n the past several years, the use of CT perfusion imaging has been rapidly growing in both the clinical and the research settings. Perfusion information about patients with stroke, chronic cerebral ischemia, vasospasm, and brain tu- mors can aid in diagnosis and treatment. CT perfusion mapping provides qualitative and quantitative information about cerebral blood flow (CBF), cerebral blood volume (CBV), and mean transit time (MTT). This information is derived from postprocessing of axial source im- ages obtained from continuous rapid scanning through a fixed level in the brain during IV in- jection of a small contrast bolus. CT perfusion maps can be postprocessed with software that has fully automated and semiautomated func- tions. The semiautomated function requires a user to select several parameters that are ap- plied in a mathematic model for generation of parametric maps. These user-defined parame- ters include arterial input function, venous function, and cutoff values for unenhanced and enhanced images. With the fully automated program, parameters are selected without input from a user. The program does, however, re- quire the user to accept or adjust each parame- ter before the CT perfusion maps are computed. The quantitative information, CBF, CBV, and MTT, can be affected by selection of these user- defined parameters. This scenario specifically applies to the CT Perfusion program (GE Healthcare) [1]. The responsibility for postpro- cessing data is held by individuals with varying levels of skill and experience, ranging from ra- diology attending physicians to CT technolo- gists. Therefore it is important to evaluate the variability that may exist in a technique that re- quires observer interaction for acquisition of data before they are interpreted. I Downloaded from www.ajronline.org by 52.73.204.196 on 05/18/22 from IP address 52.73.204.196. Copyright ARRS. For personal use only; all rights reserved