Improved deconvolution of perfusion MRI data in the presence of bolus delay and dispersion

Cerebral blood flow (CBF) is commonly estimated from the maximum of the residue function deconvolved from bolus-tracking data. The bolus may become delayed and/or dispersed in the vessels feeding the tissue, resulting in the calculation of an effective residue function, R-eff(t), whose shape reflects the distortion of the bolus as well as the hemodynamic tissue status. Consequently the CBF is often underestimated. Since regularizing the deconvolution introduces additional distortions to R-eff(t), it is impossible to distinguish a true decrease in the CBF from bias introduced by abnormal vasculature. This may result in misidentification of tissue regions at risk of infarction, which could have serious clinical consequences. We propose a modified maximum-likelihood expectation-maximization (mML-EM) method, which is shown by way of simulations to improve the characterization of R-eff(t) across a wide range of shapes. A pointwise termination approach for the iteration minimizes the effects of noise, and appropriate integral approximations minimize discretization errors. mML-EM was applied to data from a patient with left internal carotid artery (ICA) occlusion. The shape of each calculated R-eff(t) was used to create a map indicating voxels affected by bolus delay and/or dispersion, where CBF estimates are inherently unreliable. Such maps would be a useful adjunct for interpreting bolus-tracking data.