Removing the effect of SVD algorithmic artifacts present in quantitative MR perfusion studies

• DOI: 10.1002/mrm.20006
• Publication Date: 2004-02-27T08:00:36Z
• Authors: M.R. Smith; H. Lu; S. Trochet; R. Frayne
• Keywords: Gadolinium DTPA; Time Factors; Fourier Analysis; Phantoms, Imaging; Contrast Media; Reproducibility of Results; Middle Aged; Models, Theoretical; Image Enhancement; Magnetic Resonance Imaging; Sensitivity and Specificity; Brain Ischemia; Stroke; 03 medical and health sciences; 0302 clinical medicine; Cerebrovascular Circulation; Humans; Computer Simulation; Artifacts; Algorithms

AbstractQuantitative cerebral blood flow (CBF) values can be obtained from dynamic susceptibility contrast (DSC) MR perfusion studies using the standard singular value decomposition (sSVD) deconvolution algorithm. Reports in the literature from simulation and in vivo studies suggest that CBF estimates obtained using sSVD deconvolution depend on the arterial‐tissue delay (ATD). By contrast, Fourier transform (FT) deconvolution produces CBF estimates that are independent of ATD. The diagnostic reliability of quantitative CBF measurements to define areas of normal tissue flow and tissue at risk is brought into doubt by such gross sensitivity to the specifics of the deconvolution approach. This variation of CBF values with ATD is shown to be an artifact associated with the current implementation of the sSVD deconvolution algorithm. A reformulated version of the SVD deconvolution algorithm (rSVD) is presented and compared to the standard SVD algorithm through simulation and patient case studies. Magn Reson Med 51:631–634, 2004. © 2004 Wiley‐Liss, Inc.