Biomedical Engineering Reference
In-Depth Information
FIGURE 14.10
Variability Map
The images show the average intensity volume of 17 ICBM subjects mapped into stereotaxic
space nonlinearly along with the corresponding slices through the average variability map.
The cross marker (
x
14 mm) is near the planum temporale
on the left side, a region known to be variable, measured here to be 6.3 mm 3-D
FWHM
, and
appearing more variable than the right.
44 mm,
y
37 mm,
z
stereotaxic space. This identification can be achieved automatically by ANI-
MAL, since the recovered deformation field essentially establishes correspon-
dence between homologous points. The field can be interpreted as a map of
“positional differences” between individual source (after affine transformation)
and target volumes. In other words, for every 3D coordinate in the target
space, the registration procedure yields a vector-valued estimate of the dif-
ference in position between the two data sets.
This information derived from the deformation fields, and when averaged
over a large number of individual
target pairs, can yield estimates of normal
anatomical variability. The standard deviation at each voxel position (for 17
subjects from the ICBM data base) is computed separately for each of the
x
,
y,
and
z
components. These values are combined to yield a single number for
each voxel measuring intersubject variability (ISV), equivalent to a 3D
FWHM
measure. For a Gaussian distribution:
fwhm
2.35
(
2
2
2
)
/3
The 17 deformation fields were used to compute the anatomical variability
map shown in Figure 14.10. The regions of largest neuroanatomical variabil-
ity were posterior poles of the lateral ventricles, the region near the fourth
ventricle, the cingulate sulcus (slightly more on the left than the right), the
inferior frontal lobe, and the area just above the splenium of the corpus callo-
sum. The anatomical variability map is not symmetric on left and right sides.
The left frontal lobe and right parieto-occipital lobe appear to be more vari-
able than their counterparts. These data have been partially validated with
manual estimates of intersubject anatomical variability (regression coeffi-
cient of 0.867) demonstrating good correlation between both automatic and
manual methods at the 1% significance level.
97
