Biomedical Engineering Reference
In-Depth Information
abdomen has irregular boundaries, unlike the head to which registration has
been most often applied. Second, the normal prostate is a small organ that
when healthy measures only about 38.0 mm in its widest dimension trans-
versely across the base [6]. Third, different patient positions such as legs
flat and raised significantly change the legs in lower portions of image vol-
umes as well as cause movement and deformation of internal organs in the
pelvis. Fourth, the prostate might move relative to the pelvic bones due to
changes in bladder and rectal filling [7, 8]. The alignment of the pelvic bones,
a most prominent anatomical feature in MR grayscale images, does not nec-
essarily mean that the prostate is aligned. In addition, it is more difficult to
evaluate pelvic and/or prostate registration because no external markers are
available.
Many reports describe methods and evaluations for registration in the
brain [9]. A few describe results for the pelvis or prostate. For example, man-
ual registration has been used where an operator cues on segmented vascu-
lar structures [10] or other anatomical landmarks [11, 13]. Others have used
automated 3D schemes that match contours of bones and sometimes other
structures that are extracted using manual or interactive segmentation [14,
16]. Manual segmentation has also been used to create surfaces for auto-
matic registration [17, 18]. All of these methods require either segmentation
or visual identification of structures. Voxel based methods, particularly those
based upon mutual information, are robust, require no segmentation that can
be prone to error, are suitable for multimodality registration, are highly ac-
curate for brain registration [19], and are suitable for abdominal registra-
tion [20]. For registration of brain and other organs, registration accuracy
has been assessed using fiducial markers [21, 22] and anatomical landmarks
[23, 25].
The next section will describe a three-dimensional mutual information
rigid body registration algorithm with special features for MRI volumes
of the pelvis and prostate. Section 3.3 describes a three-dimensional non-
rigid registration algorithm that is based upon independent optimization
of many interactively placed control points using mutual information and
a thin plate spline transformation. Detailed implementation, comparisons
with rigid body method, and discussions are reported at the end of the
section.
