Biomedical Engineering Reference
In-Depth Information
the corpus callosum, basal ganglia, thalamus, or lateral ventricles. These
findings suggest that in the range between 5 and 18 years, the brain continues
to change its shape locally. It remains to be determined what underlies such
changes in the brain's gross anatomy.
14.3.2
Segmentation-Based Analysis of Normal Brain Anatomy
The previously described procedures do not rely on
a priori
regional identifica-
tion in each volume. Therefore, localization of the results must be interpreted
with respect to the blurred anatomy visible in the average MRI volumes in con-
junction with brain atlases registered within the same space. Region of interest
(ROI) analysis, respecting specific anatomical boundaries in each subject's
brain, requires segmentation of regions (or volumes) of interest within each
MR image volume.
Segmentation can be accomplished using manual tools that allow the user to
identify the voxels of a given structure by voxel painting using
Display
, a com-
puter program developed in our laboratory
87
that shows four 2D orthogonal
slices (transverse, coronal, sagittal, and user defined oblique) through the vol-
ume with arbitrary pan, zoom, and intensity mapping on each slice. Display
also includes a 3D graphics window capable of displaying 3D geometric
objects such as the cortical surface. The cursor can be placed in any of the 2D or
3D windows, and its position is simultaneously updated in the other views.
Voxel labels are painted on any of three orthogonal views with simultaneous
update in all other views. Within our group at the MNI, this painting technique
has been used to characterize the normal anatomy of the cingulate and para-
cingulate sulci in 105 subjects,
88
the corpus callosum in 100 subjects, the
planum temporale in 50 subjects,
90
the hippocampus and amygdala in 80 sub-
jects,
86
the lobes and fissures of the cerebellum in 12 subjects, as well as the
distribution of frontal and lateral gyri,
92
orbitofrontal gyri,
93
and the region of
the parieto-temporo-occipital cortex.
94
The manual method has also been used
to identify and then compare the frontal lobe anatomy of patients with schizo-
phrenia with age-matched control subjects.
95
Since manual labeling is prohibitively time consuming and error prone, we
have designed an automated procedure called ANIMAL (automatic nonlin-
ear image matching and anatomical labeling) to segment objectively gross
anatomical structures from 3D MRIs of normal brains.
19,96
Automatic segmen-
tation is achieved by estimating the nonlinear spatial transformation
required to register all voxels from a subject's MRI volume with an average
MRI brain that is coregistered with a SPAM atlas in a Talairach-like stereo-
taxic space.
71
The atlas' 90 average gross anatomical structures are mapped
through the inverse transform to effectively define customized masks on the
subject's MRI for the
most-likely
region for each structure. Tissue classes such
as gray matter, white matter, and CSF, identified by a minimum distance clas-
sifier, are masked by these regions to complete the segmentation. This method-
ology was applied to the collection of 152 MRI brains in the ICBM data base.
82
Since each structure has been segmented, it is possible to compute its
native volume (reported in cm
3
in Figure 14.8 and Table 14.1). Table 14.1
89
91
