Biomedical Engineering Reference
In-Depth Information
sensitivity of RAVENS with the widely used VBM approach of the SPM package
[67], and we found that RAVENS performed significantly better in this validation
study.
3.4. Longitudinal Stability
With a growing interest in longitudinal studies, which are important in study-
ing development, normal, aging, early markers of Alzheimer's disease, and re-
sponse to various treatments, amongst others, securing longitudinal stability of the
measurements is of paramount importance. However, in a longitudinal morpho-
metric study, we would typically measure the shape transformation during each
time point, and then examine longitudinal changes in the shape transformation.
This approach is valid in theory, but limited in practice. This is because small
error measurements are dramatically amplified when we calculate temporal dif-
ferences. Although temporal smoothing can be applied retrospectively to shape
measurements, it is far better if temporal smoothing is actually incorporated into
the procedure for finding the shape transformation, when the image information
is available to the algorithm, rather than retrospectively adjusting a noisy shape
transformation. The issue of longitudinal measurement robustness is particularly
important in measuring the progression of a normal older adult into mild cognitive
impairment, which makes it important to have the ability to detect subtle morpho-
logical changes well before severe cognitive decline appears. To further illustrate
the difficulties that the current 3D method is facing, in Figure 7 we have shown
some representative longitudinal volumetric measurements from single subjects
as well as from averages obtained from 90 older individuals over 6 years.
In order to address this issue and be able to obtain longitudinally stable mea-
surements, we have developed an approach to finding the shape transformation in
4D, with the 4th dimension being time [71]. The formulation is readily reduced
to a 3D problem, if only cross-sectional data are available. We should note that a
step toward this proposed direction was proposed in [37], in which the image at
one time-point was used as the template for shape reconstruction in another frame.
However, that approach still measures longitudinal differences independently for
different time-points, and therefore it does not apply temporal smoothing other
than by using the same anatomy of a different time-point as the template.
The 4Dwarping approach of [71] simultaneously establishes longitudinal cor-
respondences in the individual as well as correspondences between the template
and the individual. This is different from the 3D warping methods, which aim at
establishing only the inter-subject correspondences between the template and the
individual in a single time-point. Specifically, 4D-HAMMER uses a fully auto-
matic four-dimensional atlas matching method that constrains the smoothness in
both the spatial and temporal domains during the hierarchical atlas matching proce-
dure, thereby producing smooth and accurate estimations of longitudinal changes.
Most importantly, morphological features and matches guiding this deformation
