This article deals with the estimation of model-based parameters, such as t
he noise variance and signal components, from magnitude magnetic resonance
(MR) images. Special attention has been paid to the estimation of T-1 - and
T-2-relaxation parameters. It is shown that most of the conventional estim
ation methods, when applied to magnitude MR images, yield biased results. A
lso, it is shown how the knowledge of the proper probability density functi
on of magnitude MR data (i.e., the Rice distribution) can be exploited so a
s to avoid (or at least reduce) such systematic errors. The proposed method
is based on maximum likelihood (ML) estimation. (C) 1999 John Wiley & Sons
, Inc.