Meeting Banner
Abstract #2164

Optimal PLD Design and Maximum Likelihood CBF Estimation for Dynamic PCASL with Rician Noise

Li Zhao1, Craig H. Meyer1, 2

1Biomedical Engineering, University of Virginia, Charlottesville, VA, United States; 2Radiology, University of Virginia, Charlottesville, VA, United States

Noise in low SNR ASL images is more accurately modeled as Rician rather than Gaussian. Least squares estimation is typically used in ASL, but this results in a biased estimate with Rician noise. This work describes a new maximum likelihood (ML) estimator and an optimal post-label delay (PLD) design for dynamic ASL assuming Rician noise. To verify the performance of CBF estimation, a simulation is performed based on low SNR dynamic ASL signal. The results show that the new ML estimator provides unbiased estimation and that optimal PLD design can reduce the variance of CBF estimation significantly.

Abstract PDF

Keywords

accuracy accurate achievable achieve acquisition acquisitions added additional although approximated arises arterial assume assuming assumption assumptions averaging better biased biomedical blood bolus bound brain care classic cluster common compartment complex constrained correct cram currently delays denominator design designing designs distribution done duration dynamic efficiency efficient engineering enough estimation estimator estimators even expectation expected experiment fisher function general gives ideal illustrated important improve improvement instead kind kinetic label labeling least likelihood linear linearly little made magnitude mapping matrix maximize minimize model modified needed noise noisy normalized note operation optimal optimized parametric patients pattern performance perfusion post prefer pulsed quality question radiology rapid real reduce regardless repeated requires sampling significantly simulated simulation simulations situation slightly spaced spin squares statistical still stroke table theoretical tissue transit treat tumor typical typically unbiased unknown variance verify want yields