Biomedical Engineering Reference
In-Depth Information
4.2. Corpus Callosum Analysis
According to the reviewed literature, the deficits in the size of the corpus
callosum and its subregions in patients with autism relative to controls is well
established (see, e.g., [30]). This finding motivates the analysis presented in this
section. The main goal is to use the difference in the CC anatomy in order to
devise a classification technique between the two groups. Unlike the traditional
approaches, which are sensitive to the segmentation ouputs as well to the effect
of volume, we propose a novel classification approach based on analyzing the
displacement fields generated from the nonrigid registration of different corpus
callosums. The details of this approach are summarized below.
Due to the distortions from which the postmortem images suffer and which
made the segmentation of the CC for these data very challenging, results corre-
sponding to the savant datasets are presented in this part of the chapter. For these
data, the CCs are extracted manually from the segmented brain images.
Figure 16. 3D registration of CC datasets using our nonrigid registration technique. See
attached CD for color version.
For each group (autistic and normal), we randomly picked four CC datasets,
one of which is chosen as the reference, and the remaining ones are registered to it
using our deformable registration method (Section 3.4). During these registration
steps, a deformation field is generated for each alignment with the reference (total
of 3 for each group), as shown in Figure 18a. These deformation fields are then
averaged, and the cumulative distribution functions are computed to represent the
changes of the magnitudes of each one of these two averaged deformation fields,
as shown in Figure 18b.
Given a subject to be classified, we register it, once to the chosen control
reference and then to the chosen autistic reference. Two deformation fields are
then generated and, as we did before, a CDF is computed for each one of these
deformation fields. Finally, we compute the Levy distance between each of these
deformation fields and the average CDFs of deformation fields representing each
Search WWH ::




Custom Search