Temporal filtering of event-related fMRI data using cross-validation

To circumvent the problem of low signal-to-noise ratio (SNR) in event-related fMRI data, the fMRI experiment is typically designed to consist of repeated presentations of the stimulus and measurements of the response, allowing for subsequent averaging of the resulting data. Due to factors such as time limitation, subject motion, habituation, and fatigue, practical constraints on the number of repetitions exist. Thus, filtering is commonly applied to further improve the SNR of the averaged data. Here, a time-varying filter based on theoretical work by Nowak is employed. This filter operates under the stationary wavelet transform framework and is demonstrated to lead to good estimates of the true signals in simulated data. The utility of the filter is also shown using experimental data obtained with a visual-motor paradigm. (C) 2000 Academic Press.