Biomedical Engineering Reference
In-Depth Information
anatomy of interest falls within the scanning volume, and that the patient is
generally placed in some orientation such that gross anatomical features are
placed in some direction (for example, the patient's nose points upward in the
scanner). In the context of qualitative evaluation of MS studies however, this
arrangement leads to hindrances in evaluation, due to the fact that multiple scans
must be compared with one another. Due to the largely arbitrary positioning of
the anatomy in the scanner, in a slice-by-slice comparison between studies,
quite different anatomy can by chance be located on the same slice numbers in
different studies.
It is desirable to be able to perform registration using the intrinsic ap-
proaches, rather than imposing limitations in the scanning procedure, or affixing
artificial fiducial markers on the patient's head. For best accuracy, artificial mark-
ers would likely be affixed to the skull, and therefore would be inconvenient and
potentially painful for the patient. Additionally, this procedure would also intro-
duce a risk of infection. Furthermore, using intrinsic registration techniques, it is
also desirable to be able to apply registration retroactively, allowing for current
data sets to be aligned with data sets taken previously in a patient's history, or
perhaps with an imaging modality that prevents the use of artificial markers. In
the study of MS using MRI, for comparison of scans taken at different points of
time in a clinical study, of the same patient, a registration technique is necessary.
Such a tool would allow for alignment of the patient's anatomy in different scans.
When this alignment is accomplished, qualitative comparison of scans becomes
easier to an expert viewer, as image slices will now contain the same anatomy,
and quantitative comparison between studies is enabled in the same manner.
Registration can also be used to assist in segmentation. For example, if a model
of the patient's anatomy is known, then a study can be registered to that model,
allowing for segmentation of certain classes of problems to be made trivial, as
the segmentation of the data set is then known a priori from the model. In the
MS research, this approach is used for segmentation of a patient's brain from
his head. In this context, registration is performed by maximization of mutual
information. This technique has generally been found to perform well, and is
useful in the MS clinical settings. This technique was studied, implemented, and
tested using MS patient studies. Additionally, the performance of this method
was enhanced by application of parallel programming techniques.
Figure 1.15 shows sample slices from the registration studies. Each row of
the figure corresponds to a single study. The first column of the figure shows a
sample slice from the floating volume used for each study. The second column of
