Image Processing Reference
In-Depth Information
parameters of the T1-weighted 3-D measurement. Again, the underlying assumption
is that subjects do not move between the functional and anatomical scans. However,
because small head movements cannot be excluded, the accuracy of the coregistra-
tion obtained in this way by means of various landmark-based, edge-matching or
other automatic registration algorithms. These algorithms produce a spatial transfor-
mation that ensures an accurate coregistration between anatomical and functional
data. By applying this spatial transformation to each volume of the functional time
series, it is possible to generate a new 4-D representation of the functional data in
which the functional time courses are directly linked to the anatomical reference
volume. This 4-D representation also makes possible (after spatial normalization
[see Subsection 15.6.2 ]) the concatenation of the time series of multiple subjects for
a group analysis, e.g., within the GLM framework.
15.6.2
S PATIAL N ORMALIZATION
The comparison of spatial locations of functional activation among subjects is
commonly made in both fMRI and PET studies by normalizing the individual
brains in a standard space. The most widely used standard anatomical reference
is the stereotaxic space, which was defined in the Talairach and Tournoux atlas
[17]. In our approach, the transformation in Talairach space is performed semi-
automatically using the 3-D anatomical volume of each subject and following
the procedure defined in the atlas:
1.
In the first step, the 3-D anatomical data set of each subject is rotated
in order to align it with the stereotaxic axes. For this step, the location
of the anterior commissure (AC) and the posterior commissure (PC)
and the two rotation parameters for midsagittal alignment have to be
specified manually in the 3-D data set.
2.
In the second step, the extreme points of the cerebrum (anterior, pos-
terior, superior, inferior, left and right) are specified. Planes encapsu-
lating these points together with the vertical frontal plane (VAC, the
plane established along the AC and bisecting the AC-PC line orthog-
onally) and the vertical posterior plane (VPC, the plane established
along the PC and bisecting the AC-PC line orthogonally) divide the
brain into 12 subvolumes.
3.
The 3-D data sets are scaled into the dimensions defined in the Talairach
and Tournoux [17] atlas by applying separately to each of the 12
subvolumes a piecewise affine and continuous transformation.
Note that normalization in the volumetric Talairach space only ensures a
coarse spatial correspondence between brains of different subjects. More
advanced algorithms, based on the realignment of the subjects' cortices (see the
following section), have been recently developed to address the problem of
defining a spatial correspondence between different brains, which is the basis of
all intersubject comparisons and group analyses in functional neuroimaging.
Search WWH ::




Custom Search