Myocardial blood flow quantification with MRI by model-independent deconvolution

Michael Jerosch-Herold, Cory Swingen, Ravi Teja Seethamraju

Research output: Contribution to journalArticle

206 Scopus citations

Abstract

Magnetic resonance (MR) imaging during the first pass of an injected contrast agent has been used to assess myocardial perfusion, but the quantification of blood flow has been generally judged as too complex for its clinical application. This study demonstrates the feasibility of applying model-independent deconvolution to the measured tissue residue curves to quantify myocardial perfusion. Model-independent approaches only require minimal user interaction or expertise in modeling. Monte Carlo simulations were performed with contrast-to-noise ratios typical of MR myocardial perfusion studies to determine the accuracy of the resulting blood flow estimates. With a B-spline representation of the tissue impulse response and Tikhonov regularization, the bias of blood flow estimates obtained by model-independent deconvolution was less than 1% in all cases for peak contrast to noise ratios in the range from 15:1 to 20:1. The relative dispersion of blood flow estimates in Monte Carlo simulations was less than 7%. Comparison of MR blood flow estimates against measurements with radio-isotope labeled microspheres indicated excellent linear correlation (R2 = 0.995, slope: 0.96, intercept: 0.06). It can be concluded from these studies that the application of myocardial blood flow quantification with MRI can be performed with model-independent methods, and this should support a more widespread use of blood flow quantification in the clinical environment.

Original languageEnglish (US)
Pages (from-to)886-897
Number of pages12
JournalMedical Physics
Volume29
Issue number5
DOIs
StatePublished - Jan 1 2002

Keywords

  • Magnetic resonance imaging
  • Model-independent quantification
  • Myocardial blood flow
  • Tikhonov regularization

Fingerprint Dive into the research topics of 'Myocardial blood flow quantification with MRI by model-independent deconvolution'. Together they form a unique fingerprint.

  • Cite this