Fusion of clinical information for segmentation purposes:
Application to prostate segmentation for curietherapy
Geoffroy Rivet-Sabourin
Denis Laurendeau (Supervisor)
Problem: Prostate cancer is the cancer which affects the greatest number of Canadian men. For many years, research has been devoted to developing the earliest possible detection methods for this type of cancer as well as less and less invasive treatment procedures. It is within this context that prostate cancer treatment using curietherapy is investigated. This method consists in implanting permanent radioactive sources inside the prostate.
Approach: The clinical protocol consists in taking a series of ultrasonography images so as to determine the volume of the prostate. A computerized system will then use this volume to calculate the optimum positions for the radioactive sources to enable the entire prostate to be irradiated, while protecting the organs at risk (bladder and rectum). The patient is then admitted to the operating block to have the sources implanted. The sources are implanted under ultrasonographic guidance. Before leaving the operating block, the radio-oncologist traces over the contours of the prostate on the ultrasonography one last time. Then 30 days later, the radio-oncologist uses the computerized tomography (CT) images to validate the position of the sources and the quality of the treatment. The difficulty of this last step is the precision with which the radio-oncologist can identify the prostate contour. The literature and our experiments indicate an inter-operator variation of 30% on the volume generated by the contours of the doctor.

The method proposed consists in using the information gathered during the clinical protocol (contours on the ultrasonography images, positions of the sources) so as to produce a series of initial contours on the CT images 30 days after they have been implanted. The method is initially based on the positions of the sources implanted in the prostate to align the ultrasonography contours on the CT images. The initial contours are deformed image by image based on the positions of the sources and based on the image gradient (active contour). This involves a series of deformations by active contours, where each deformation step is followed by a step involving the smoothing of the 3D surface.

The algorithm produces a series of contours which are as close as possible to the final contour which will be traced by the doctor. A clinical study is currently underway to validate the algorithm. A data base of twelve patients who have been subjected to prostate curietherapy has been established. These patients were referred to five doctors. This study will be able to evaluate if the use of the algorithm enables a reduction in both the intra- and inter-operator variability.

Last modification: Nov 19 2007 10:24AM by grivet


©2002-. Computer Vision and Systems Laboratory. All rights reserved