Software Packages for Data Procesing

WRST Analysis Toolbox

The Wavelet Regularized SpatioTemporal (WRST) Analysis toolbox for fMRI comprises a set of matlab-based programs that implement some of our recently developed analysis methods for single subject fMRI data. The WRST toolbox fits a physiological model for the BOLD signal under appropriate noise assumptions to the motion corrected fMRI data, and estimates the noise processes at each voxel using an EM algorithm. WRST merges these voxel-wise estimates into a spatial estimation procedure that uses a modified wavelet thresholding to better preserve spatial heterogeneity in the activation maps while providing the necessary regularization. Thus the popular approach of spatially presmoothing the fMRI data with a fixed-width Gaussian kernel is no longer necessary, as all spatial operations are treated within a full spatiotemporal estimation. The wavelet estimation is nonstandard in this regard, since, unlike typical spatial wavelet techniques, it is modulated by the temporal characteristics of the data, providing better spatial adaptivity of the functional clusters.

The user may if desired, compare WRST performance with either standard presmoothing, or a similar strategy that replaces the wavelet step with Gaussian smoothing. i.e. the signal parameter estimates are smoothed directly as opposed to smoothing the raw data (presmoothing). This latter choice allows one to directly determine the benefits of adding the wavelet smoothing in the overall procedure. For completeness, the program also includes an option that allows replacement of the modified wavelet procedure, distinct to WRST, with a standard spatial wavelet smoother. This enables the user to contrast between a spatial wavelet + temporal approach and the full spatiotemporal wavelet analysis.