Pixel-wise partial volume effects correction on arterial spin labeling magnetic resonance images

Abstract

Arterial spin labeling is a recently emerging imaging modality in functional magnetic resonance, and it is widely acknowledged to be effective in directly measuring the cerebral blood flow of patients while being scanned, which makes it a promising indicator in contemporary dementia disease diagnosis studies. However, partial volume effects mainly caused by signal cross-contamination due to pixel heterogeneity and limited spatial resolution of the arterial spin labeling scanning protocol often prevent the cerebral blood flow from being accurately measured. In order to correct the partial volume effects, contemporary studies usually rely on neighboring pixles to solve indefinite equations of partial volume correction, which makes shortcomings of blurring and brain tissue details loss inevitable in their correction outcomes. In this study, a novel pixel-wise correction method is proposed to tackle partial volume effects in arterial spin labeling images. The main idea is to formalize the correction problem as a series of quadratic programming sub-problems using split-Bregman iterations, then formulate each sub-problem via the regularization of least absolute shrinkage and selection operator, and finally solve the regularization sub-problem via the fast proximal gradient descent approach. A real-patients database composed of 360 demented patients is incorporated for experimental evaluation of the pixel-wise method. Extensive experiments and comprehensive statistical analysis are carried out to demonstrate the superiority of the pixel-wise method with comparisons towards the popular region-based method. Promising results are reported from the statistical point of view.