Adaptive Smoothing Applied to fMRI Data

Abstract

One problem of fMRI images is that they include some noise coming from many other sources like the heart beat, breathing and head motion artifacts. All these sources degrade the data and can cause wrong results in the statistical analysis. In order to reduce as much as possible the amount of noise and to improve signal detection, the fMRI data is spatially smoothed prior to the analysis. The most common and standardized method to do this task is by using a Gaussian filter. The principal problem of this method is that some regions may be under-smoothed, while others may be over-smoothed. This is caused by the fact that the extent of smoothing is chosen independently of the data and is assumed to be equal across the image. To avoid these problems, we suggest in our work to use an adaptive Wiener filter which smooths the images adaptively, performing a little smoothing where variance is large and more smoothing where the variance is small. In general, the results that we obtained with the adaptive filter are better than those obtained with the Gaussian kernel. In this paper we compare the effects of the smoothing with a Gaussian kernel and with an adaptive Wiener filter, in order to demonstrate the benefits of the proposed approach.