The simplest practical form of spatial normalization is the
piecewise linear Talairach transformation which has been discussed
in Section II-A. Because the Talairach coordinate system
has been widely used and is generally accepted as the standard
stereotaxic reference, it is normally used for postanalysis
activation labeling, and the Talairach transformation has been
used in many group-analysis studies. Nevertheless, after Talairach
registration, differences on the order of centimeters may
be observed between the anatomical landmarks detected in the
spatially normalized brains [192]. The fact that the low-dimensional
Talairach transformation cannot adequately deal with the
anatomical variability in different brains has been the motivation
for developing higher dimensional spatial normalization techniques.
Even other low-dimensional warping techniques, such
as polynomial warping [193], have been shown to be more accurate
than the Talairach transformation for intersubject anatomical
landmark matching [194]. The other drawback of the Talairach
transformation is that it is based on manual selection of
landmark anatomic reference points.
The accuracy of spatial normalization along