Three-Dimensional Rigid and Non-Rigid Image Registration for the Pelvis and Prostate - Biomedical Image Analysis - page 111

over a 219 × 350 × 192-mm rectangular FOV and 1 . 37 × 1 . 37 × 1 . 5-mm voxels
oriented to give the highest resolution for transverse slices. There was oversam-
pling at 31% in the slice direction to reduce aliasing artifacts. The acquisition
time was 4.25 minutes. Most often, we used the second sequence, which gave
excellent image contrast for the prostate and its surroundings.

3.2.4.2 Image Volumes for Registration

We acquired 3D MRI volume images from three prostate cancer patients and
three normal volunteers under four conditions simulating anticipated situa-
tions in diagnostic and treatment applications. They are diagnostic position ,
treatment position , empty bladder , and diagnosis 1 week . In the diagnostic
position , the subject laid supine throughout MR scanning. The reference volume
was always obtained in the diagnostic position. In the treatment position , the
subject was supine, and his legs were supported at 30 ◦ -60 ◦ relative to the hori-
zontal and separated in a “V” with an angle of 60 ◦ -90 ◦ between the legs. This is
similar to the lithotomy position used in some prostate therapies, and it should
provide access for needle insertion in brachytherapy or RF thermal ablation. In
some experiments, the subject micturated to create an empty bladder prior to
imaging. For each subject, volumes were typically obtained within an imaging
session of 1-2 hours. We imaged one subject (V2) a week before the standard
imaging session, and we refer to this volume as diagnosis 1 week. Between vol-
ume acquisitions, subjects got off the MRI table, stretched, and walked around
to ensure that they would assume a different position on the table. The coil array
was centered on the prostate. All images of a subject were acquired using the
same pulse sequence and acquisition parameters so as to ensure very similar
gray values. In total, we registered 22 volume pairs consisting of one pair for each
patient, six pairs for each volunteer, and one additional pair for volunteer V2.

3.2.4.3 Assessments of Pelvic Registration

Following registration, we determined displacements between the six bony land-
marks. For each subject, there was no consistent displacement of landmarks in
one direction versus another. Hence, we measured 3D distances and determined
RMS values over the six landmarks. Registration results are plotted in Fig. 3.4.
The smallest errors are obtained when subjects are in the diagnostic position for