Most partial Fourier (PF) approaches use a low-resolution phase estimate in
the reconstruction to account for non-zero phases of the image. These meth
ods may fail when there are large phase errors, a situation commonly encoun
tered in T-2(*)-weighted functional MRI (fMRI). To mitigate this problem, a
method was developed based on the inversion of a matrix formulated accordi
ng to a phase map derived from iterative reconstruction. To make this metho
d computationally practical for fMRI, a strategy was introduced such that t
he matrix inversion is performed only once for each slice in the time serie
s, assuming that the phase map remains constant in the time series. To ensu
re the temporal phase invariance, physiological noise correction and global
phase correction were applied to the data before the reconstruction. This
method was demonstrated to be robust and efficient for fMRI. (C) 2001 Wiley
-Liss, Inc.