Smoothing FMRI Data Using an Adaptive Wiener Filter

Abstract

The analysis of fMRI allows mapping the brain and identifying brain regions activated by a particular task. Prior to the analysis, several steps are carried out to prepare the data. One of these is the spatial smoothing whose aim is to eliminate the noise which can cause errors in the analysis. The most common method to perform this is by using a Gaussian filter, in which the extent of smoothing is assumed to be equal across the image. As a result some regions may be under-smoothed, while others may be over-smoothed. Thus, we suggest smoothing the images adaptively using a Wiener filter which allows varying the extent of smoothing according to the changing characteristics of the image. Therefore, we compared the effects of the smoothing with a wiener filter and with a Gaussian Kernel. In general, the results obtained with the adaptive filter were better than those obtained with the Gaussian filter.