Extraction of the hemodynamic response in randomized event-related functional MRI

Signal detection in a noisy data set is a common problem in signal processing. Detection of the hemodynamic response function (HRF) embedded in randomized event-related fMRI (rER-fMRI) time series is an example of this problem. So far, most studies that set out to obtain unbiased HRF use some forms of time-window (TW) averaging method to extract HRF from the rER-fMRI data. In this paper we applied two methods, cepstral analysis and conjugate gradients (CG) for deconvolution to extract HRF. These methods depend only on the knowledge of when events occurred and do not require any a priori information about the HRF. These methods and the popular TW averaging method are tested on simulated data, as well as in vivo data obtained from rER-fMRI experiments. All three methods identified timing of HRF accurately, but only the CG for deconvolution method was robust in reproducing the shape under varying experimental conditions.