Encrypted login | home

Program Information

How Accurate Do Prostate Contours Have to Be?

D Wiant

D Wiant*, B Sintay , J Maurer , T Hayes , Q Shang , H Liu , Cone Health Cancer Center, Greensboro, NC


WE-RAM1-GePD-JT-4 (Wednesday, August 2, 2017) 9:30 AM - 10:00 AM Room: Joint Imaging-Therapy ePoster Theater

Purpose: Automated contouring offers improvements in consistency and time. However, few centers have adopted automated techniques, often citing contour accuracy concerns. In this work, we examine the relationship of organ / target shape change over the course of treatment with the level of contouring accuracy needed to adequately model delivered dose.

Methods: Daily cone beam computed tomography scans (CBCT) were retrospectively reviewed for 4 prostate patients that received 40 fractions. Prostate, bladder, and rectum were contoured on each CBCT by a physicist then reviewed by a second physicist. All scans were rigidly registered based on bony anatomy to the day 1 CBCT in MIM Maestro (MIM Software, Cleveland, OH). The scans were than aligned based on prostate centroid. The percent overlap between the day 1 and following scans along with the expansion of the day 1 contours needed to cover the following day contours up to 9 mm were recorded.

Results: The mean overlap for the prostate over all patients and fractions was 0.87 +/- 0.07 (0.66-1). The overlap for the bladder was 0.81 +/- 0.17 (0.26-1). The overlap for the rectum was 0.84 +/- 0.10 (0.61-0.99). The mean expansion to cover the prostate over all patients and fractions was 5.3 +/- 1.8 mm (2-9 mm). The expansion to cover the bladder was 6.6 +/- 2.9 mm (0-9 mm). The expansion to cover the rectum was 6.5 +/- 2.0 mm (2-9 mm).

Conclusion: Even after allowing for inter-user contour variability and the small sample size the prostate shows deformation and the other organs clearly show deformation / displacement relative to the prostate on a daily basis. The initial contours do not represent the anatomy with sub-millimeter accuracy over the course of treatment. With this in mind automated segmentation or the use of auto-generated control structures may be feasible.

Contact Email: