Respiratory Motion Prediction with Extended Kalman Filters Based on Local Circular Motion Model

Motion of thoracic and abdominal tumors induced by respiratory motion often exceeds more than one centimeter which can compromise dose conformality significantly. Motionadaptive radiotherapy aims to deliver a conformal dose distribution to a tumor with minimal normal tissue exposure, by compensating for the tumor motion in real time. This requires prediction of respiratory motion to estimate the respiratory movement that has occurred during the system latency due to measurement and control. One of the most successful models for predicting respiratory motion is the local circular motion (LCM) model. It characterizes the local respiratory behavior with a circular motion in an augmented plane and captures the natural evolution of respiratory motion. In this paper, we utilize the first and second-order extended Kalman filters based on LCM model for predicting respiratory motion. We also optimize the parameters of the extended Kalman filters for each trace in an attempt to improve prediction accuracy. Numerical experiments are performed to evaluate and compare prediction accuracy of four different prediction schemes.