Biomedical Engineering Reference
In-Depth Information
or darkening of an MR image due to demyelination, deposition of minerals, or
other macro- or microstructural changes caused by disease. Vascular disease also
causes well-known MR signal changes, for example in the white matter of the
brain (e.g., brightening of a T
2
-weighted signal). It is thus becoming clear that
multiple modalities and multiple anatomical regions must be considered jointly in
a multivariate classification fashion, in order to achieve the desirable diagnostic
power. Moreover, regions that are relatively less affected by disease should also
be considered along with regions known to be affected (which for the example of
Alzheimer's Diseasemight include primarily temporal lobe structures, in relatively
early disease stages), since differential atrophy or image intensity changes between
these regions are likely to further amplify diagnostic accuracy and discrimination
from a background of normal variation.
The approach described in [72] is based on the RAVENS mass-preserving
morphological representation described earlier in this chapter. It hierarchically
decomposes a RAVENS map into images of different scales, each capturing the
morphology of the anatomy of interest at a different degree of spatial resolution.
The most important morphological parameters are then selected and used in con-
junction with a nonlinear pattern classification technique to form a hypersurface,
the high-dimensional analog to a surface, which is constructed in a way that it op-
timally separates two groups of interest, for example, normal controls and patients
of a particular disease. Effectively, that approach defines a nonlinear combination
of a large number of volumetric measurements from the entire brain, each taken
at a different scale that typically depends on the size of the respective anatomi-
cal structure and the size of the region that is most affected by the disease. This
nonlinear combination of volumetric measurements is the best way to distinguish
between the two groups, and therefore to perform diagnosis via classification of a
new scan into patients or normal controls.
3.6. Neuroimaging Studies of Aging, Schizophrenia, and Genetic
Influences on Brain Development
Voxel-based morphometric analysis has been adopted in a variety of stud-
ies. Here, we briefly summarize three studies in which we have applied these
techniques, in order to illustrate their use.
3.6.1. Baltimore Longitudinal Study of Aging(BLSA)
The neuroimaging arm of the BLSA was initiated in 1993, and it is now in
its 11th year [3, 76]. Approximately 150 healthy older adults have been followed
annually over this period with structural, functional (PET-
15
O), and neuropsycho-
logical evaluations. Analysis and integration of these data aims at determining
early markers of Alzheimer's disease (AD) in a background of structural and func-
tional changes occurring in normal aging. The RAVENS methodology described
