Segmenting the male pelvic organs from limited angle images with application to ART Public Deposited

Downloadable Content

Download PDF
Last Modified
  • March 21, 2019
  • Frederick, Charles Brandon
    • Affiliation: School of Medicine, UNC/NCSU Joint Department of Biomedical Engineering
  • Prostate cancer is the second leading cause of cancer deaths in men, and external beam radiotherapy is a common method for treating prostate cancer. In a clinically state-of-the-art radiotherapy protocol, CT images are taken at treatment time and are used to properly position the patient with respect to the treatment device. In adaptive radiotherapy (ART), this image is used to approximate the actual radiation dose delivered to the patient and track the progress of therapy. Doing so, however, requires that the male pelvic organs of interest be segmented and that correspondence be established between the images (registration), such that cumulative delivered dose can be accumulated in a reference coordinate system. Because a typical prostate radiotherapy treatment is delivered over 30-40 daily fractions, there is a large non-therapeutic radiation dose delivered to the patient from daily imaging. In the interest of reducing this dose, gantry mounted limited angle imaging devices have been developed which reduce dose at the expense of image quality. However, in the male pelvis, such limited angle images are not suitable for the ART process using traditional methods. In this work, a patient specific deformation model is developed that is sufficient for use with limited angle images. This model is learned from daily CT images taken during the first several treatment fractions. Limited angle imaging can then be used for the remaining fractions at decreased dose. When the parameters of this model are set, it provides segmentation of the prostate, bladder, and rectum, correspondence between the images, and a CT-like image that can be used for dose accumulation. However, intra-patient deformation in the male pelvis is complex and quality deformation models cannot be developed from a reasonable number of training images using traditional methods. This work solves this issue by partitioning the deformation to be explained into independent sub-models that explain deformation due to articulation, deformation near to the skin, deformation of the prostate bladder, and rectum, and any residual deformation. It is demonstrated that a model that segments the prostate with accuracy comparable to inter-expert variation can be developed from 16 daily images.
Date of publication
Resource type
Rights statement
  • In Copyright
  • Lalush, David Scott
  • Doctor of Philosophy
Graduation year
  • 2013

This work has no parents.