The three-dimensional (3D) trajectory of an implanted tumor marker can be estimated from its projected 2D trajectory in a set of cone-beam CT (CBCT) projections by a probability-based method. The uncertainty in the position estimation depends on the trajectory and varies along a given trajectory. The mathematical formalism of the method includes an individualized measure of the position estimation error in terms of an estimated 1D Gaussian distribution for the unresolved target position. The present study investigates how well this 1D Gaussian predicts the actual distribution of position estimation errors. Over 5000 CBCT acquisitions were simulated from a 46-patient thoracic/abdominal and a 17-patient prostate tumor motion database. The 1D Gaussian predicted the actual root-mean-square and 95th percentile of the position estimation error with mean errors ≤0.04mm and maximum errors ≤0.48mm. This finding indicates that individualized root-mean-square errors and 95% confidence intervals can be applied reliably to the estimated target trajectories.
Proceedings of the Xvith International Conference on the Use of Computers in Radiation Therapy, 2010