Analysis and classification of fMR time series using map blind deconvolution and fourier wavelet regularized deconvolution

Abstract

The procedure to estimate brain activity based on fMR signals is a process based on many assumptions. Some of the methods such as GLM (General Linear Model) and ICA(Independent Component Analysis) used for this purpose contain several restrictions. In GLM, it is assumed that each active voxel responds similarly and linearly towards a given stimulus. In ICA, an unsurmountable number of independent time series are produced, one of which is assumed to reflect the activity pattern. In this study, we used minimal number of assumptions to estimate an underlying HRF (hemodynamic response function) from a given fMR time series, and then used the estimated HRFs to classify voxels as active or passive. We have investigated results from simulations and real fMR experiments.