Wavelet Transform Denoising for fMRI Analysis

Existence of significant noise and artifacts in the fMRI signal complicates the problem of activation detection in the time domain. Because of poor signal-tonoise ratio (SNR) of the fMRI time series and confounding effects, the results of fMRI analysis are often unsatisfactory. In addition, the structure of fMRI noise is not known and still is an open problem. This makes the fMRI noise suppression a challenging problem. Different parametric denoising methods such as wavelet based denoising methods have been introduced in the literature. But these general denoising methods are based on some assumptions, such as Gaussian noise, which may not necessarily hold for the fMRI data. To remedy this problem, using randomization analysis, we propose a novel model-free method for noise suppression in fMRI analysis. The proposed denoising method is employed in conjunction with a feature space for fMRI cluster analysis and its efficiency is shown using simulated and experimental datasets.