3D Bayesian Tracking with a Single Imager for Real-Time Image Guidance in Prostate Radiation Therapy

Intra-fraction internal anatomy motion is one of the major causes of the uncertainty in prostate radiation therapy. Real-time tracking of the intra-fraction prostate motion during radiation therapy is necessary to truly benefit from the highly conformal dose distribution. With the widespread use of kV on-board imaging devices, it is desirable to estimate the 3D tumor position from a single x-ray imager during treatment delivery. In this work, we will present an improved real-time 3D Bayesian tracking algorithm with an online update scheme. By incorporating all previously acquired images during dose delivery and updating the motion probability density function in an online fashion, the algorithm is able to track large and abrupt changes that are typical of prostate motion. The flexible Bayesian formulation allows one to easily incorporate this information and obtain the 3D motion with minimal computational cost. The new tracking algorithm has been tested on actual prostate trajectories recorded with implanted electromagnetic transponders for a total of 10 patients. The new algorithm outperformed a previous one with a fixed prior built from the original setup images (Li et al. 2011). It was found that with the new algorithm, the mean 3D tracking error is about 0.15 mm and the 95th percentile error is about 0.45 mm on average for all the patients. The proposed 3D tracking algorithm is useful for real-time image guidance in prostate radiation therapy.