Biomedical Engineering Reference
In-Depth Information
Figure 3.6: The prostate overlap between reference and registered images.
Following registration, the prostate was manually segmented in reference (a)
and diagnosis (b) images. The rectangular region in (b) is zoomed in (c) with
both boundaries superimposed. Images are from volunteer V2.
of prostate cancer treatments such as brachytherapy, chemo- or radiation
therapy.
3.2.4.5
Effects of Image Cropping
In Figure 3.7, we plot registration error as determined from bony landmarks
with and without the cropping operation described in Methods. For the case
treatment-reference, cropping always improved registration accuracy, and for
V3, error reduced greatly from 12.6 mm to 3.4 mm. For all other cases, subjects
were always in the supine position with legs flat on the table, and there was
no consistent effect of cropping. If anything, cropping tends to increase error
in these cases, with an increase in 5 of 7 volume pairs. Correlation coefficient
always improves with image cropping.
