Bayesian fisher information criterion for sampling optimization in ASL-MRI

Pulsed Arterial Spin Labeling (PASL) techniques potentially allow the absolute, non-invasive quantification of brain perfusion using Magnetic Resonance Imaging (MRI). This can be achieved by fitting a kinetic model to the data acquired at a number of inversion times (TI). Some model parameters such as the arterial transit time need to be estimated together with perfusion, while others are usually assumed to be known. The accuracy of the model estimation strongly depends on the distribution of the TI sampling points. Here, we propose a Bayesian framework for PASL perfusion estimation based on the Fisher information criterion, whereby the optimal sampling points can be determined taking into account the uncertainty of the model parameters as well as the amount of noise in the data. We show that the optimal sampling strategy for PASL depends on the a priori knowledge of the model parameters and this should therefore be taken into account.