Estimation of FMRI response delays

We present an efficient algorithm using the Hilbert Transform for estimating the delay of the BOLD response to neuronal stimulation. With minimal additional computations, the algorithm estimates parameters generated in the widely used cross-correlation method and simplifies the interpolation required to estimate the response delay from the cross-correlation function. We examined errors in the Hilbert-based delay estimate associated with the use of DFT on short-duration discrete signals and proposed a method for minimizing these errors. Furthermore, we compared the delay estimates obtained with the Hilbert method to those obtained using the onset of the BOLD response. The Hilbert method resulted in less variance in the delay estimate despite the potential for higher variability in the latter part of the BOLD response. This improved delay estimate was attributed to the reduced sensitivity of the Hilbert method to noise contamination compared to the onset method. (C) 2003 Elsevier Science (USA). All rights reserved.