Prostate positioning errors associated with two automatic registration based image guidance strategies

Daily image guidance for helical tomotherapy prostate patients is based on the registration of pre-treatment megavoltage CT (MVCT) images and the original planning CT. The goal of registration, whether manual or automatic, is the overlap of the prostate; otherwise prostate misplacement may compromise the efficacy of treatment or lead to increased toxicity. A previous study demonstrated that without the aid of implanted fiducials, manual registration results in inaccurate prostate positioning. The objective of this work is to quantify prostate misplacement that results from automatic bone matching (BM) and image matching (IM) registration algorithms. 204 MVCT images from 8 high risk tomotherapy prostate patients were incorporated into this retrospective study. BM and IM registration algorithms--based on maximization of mutual information of bony anatomy only and the entire image, respectively--were used to independently register MVCT images to their respective planning images. A correlation coefficient based algorithm that uses known planning CT contour information was used for automatic prostate localization in each MVCT image. Daily prostate misplacement was determined by repositioning as calculated from the BM and the IM algorithms. Mean (+/- SD) and maximum 3D prostate positioning errors were 3.7 +/- 2.1 mm and 11.8 mm for bone matching and 4.6 +/- 2.3 mm and 11.5 mm for image matching. In terms of translational directions, IM would lead to prostate positioning error > or = 3 mm in any of the LR, AP or SI directions in 62% of treatment fractions. The corresponding value for BM is 51%. The values for positioning errors > or = 5 mm were 29% and 17% for IM and BM, respectively. This data suggests automatic daily image guidance for tomotherapy prostate patients should be based on bone matching instead of image matching.