An expectation-maximization approach for partial volume estimation of arterial spin labeled MRI data: A feasibility study

In this paper we introduce a novel expectation maximization (EM) based partial volume estimation method for arterial spin labeling (ASL) perfusion magnetic resonance (MR) imaging. Compared with structural MR images, perfusion MR images are usually contaminated with more severe noises and have a modest spatial resolution for tissue differentiation. The proposed EM-based 4D parameter estimation approach has its advantage to adequately model the underlying statistical distribution of ASL perfusion signal and to simultaneously estimate the mixture contributions of each tissue type of interest for each voxel in the 3D spatial domain while considering the series of observations such voxel in the temporal domain. Meanwhile, the well-established maximum-a-posteriori (MAP) principle is incorporated into the EM framework, where the prior distribution of tissue mixtures is described by the Markov random field (MRF) model. The feasibility of the proposed 4D MAP-EM estimation approach was investigated by estimating the individual contribution of grey matter (GM) or white matter (WM) to the ASL perfusion in the voxel level of the brain ASL data. Experimental results demonstrated that the blood flow pattern across the brain can be sufficiently visualized by the voxel-wise tissue mixtures, which is promising for the diagnosis of various brain diseases.