Biomedical Engineering Reference
In-Depth Information
FIGURE 14.5
SPAMs of white matter, gray matter, and CSF automatically derived from 151 healthy adults.
created by voxelwise averaging of the 152 classified volumes. In these images,
the voxel intensity is directly proportional to the probability of the given tis-
sue type at that voxel location. These tissue SPAMs can therefore be used as
priors for automated (Bayesian, among others) classification methods.
The classified data have been used to study structural asymmetries in the
human brain
84
where the binary mask for each tissue class was smoothed
with a 10 mm full width at half maximum (FWHM) Gaussian filter to produce
a
tissue density map.
These maps were then flipped about the
x
0 axis (i.e.,
the longitudinal fissure) and the mirror images were subtracted from the
original map. After correcting for multiple comparisons, a voxelwise t-test
analysis of the difference images confirmed known existing asymmetries for
the planum temporale and for frontal and occipital petalias. The technique
has also demonstrated interesting differences for other regions such as the
head of the caudate nucleus (
r
l
).
To study brain development we have used voxelwise regression analysis
to investigate relationships between structural features and a combination
of independent variables. Three examples are described here. In the first, we
were interested in characterizing age-related changes in local white matter
signal throughout the brain using an MRI data base of children and adolescents
aged 4 to 17 years (66 boys, age 10.7
3.7 years).
85
The significance of the relation between age and white matter density was
assessed for each voxel in the volume by means of simple linear regression.
T-values were computed by dividing the voxel slope estimate by its standard
deviation. The presence of regions that were significantly correlated with
age was determined using the 3D Gaussian random field theory developed
by Worsley.
35
This analysis showed significant regions in the internal capsule
(see Figure 14.6) and the posterior portion of the left arcuate fasciculus.
In the second example of regression analysis, we were interested in the
anatomy of the hippocampus. Eighty subjects were selected from the ICBM
data base (41 females, 39 males, 25
3.8 years; 45 girls, age 11.5
4.9 years old). Their hippocampi
were manually segmented and averaged to create hippocampal SPAMs.
86
Even though it is generally believed that the age-related decline in hippocam-
pal volume occurs in late adulthood and is independent of gender, we found
