Biomedical Engineering Reference
In-Depth Information
FIGURE 14.6
Voxelwise regression analysis: The maps of t-statistic values are superimposed on an axial
MR section from a single subject. The image depicts the exact locations in the internal capsule
(indicated by the arrow) that showed statistically significant correlations between white
matter density and the subject's age.
a significant negative correlation with age for both left and right hippocam-
pal volume in men (
r
−
.46 and
r
−
.43, respectively) but not in women
(
r
.02, respectively) in this group of young adults. Voxel-based
regression analysis was used to determine which part of the hippocampus is
correlated with age. Here, the subject's age acts as independent variable and
the MR image signal intensity of each voxel as dependent variable in the
regression. After statistical analysis (as described above), the regression anal-
ysis revealed that the volume loss occurred mostly in the head and tail of the
hippocampus (see Color Figure 14.7*).
The last example does not use image data in the regression analysis.
Instead, age is correlated with deformation vectors (a byproduct of a nonlin-
ear registration procedure). Previous work has suggested that the size of the
brain changes little after the age of five years.
72
We sought to address the pos-
sibility that after this age the human brain continues to grow, albeit in a region-
specific fashion.
82
Each of the brains in the children
.01 and
r
adolescent MRI data base
described above was registered with a nonlinear transformation to match a
single “template” volume resident in stereotaxic space. The deformation
fields represent the local difference at each voxel between the subject and
template, and as such are indicators of individual anatomical variability.
These fields were used in voxelwise linear regression analysis to identify
regions in which age correlated with a local change in X, Y, and
or Z vectors.
Hotelling's F statistic was employed to evaluate statistical significance of
such correlations; F
9 was deemed significant after correcting for multiple
comparisons. Significant correlations (30
60) between age and the local
vectors around both the temporal lobes and the ventral aspects of the frontal
lobes were found. Analysis of the X, Y, and Z vectors indicates a possible
overall growth of the temporal lobes and
F
or their downward and lateral
“movement.” In contrast, no significant correlations were observed around
* Color Figures follow page 22.
