Reduction of errors in ASL cerebral perfusion and arterial transit time maps using image de-noising

In this work, the performance of image de-noising techniques for reducing errors in arterial spin labeling cerebral blood flow and arterial transit time estimates is investigated. Simulations were used to show that the established arterial spin labeling cerebral blood flow quantification method exhibits the bias behavior common to nonlinear model estimates, and as a result, the reduction of random errors using image de-noising can improve accuracy. To assess the effect on precision, multiple arterial spin labeling data sets acquired from the rat brain were processed using a variety of common de-noising methods (Wiener filter, anisotropic diffusion filter, gaussian filter, wavelet decomposition, and independent component analyses). The various de-noising schemes were also applied to human arterial spin labeling data to assess the possible extent of structure degradation due to excessive spatial smoothing. The animal experiments and simulated data show that noise reduction methods can suppress both random and systematic errors, improving both the precision and accuracy of cerebral blood flow measurements and the precision of transit time maps. A number of these methods (and particularly independent component analysis) were shown to achieve this aim without compromising image contrast.