Prediction of respiratory motion using a modified Recursive Least Squares algorithm

In robotic radiosurgery, a robot-mounted LINAC follows the motion of the target region. Due to delays in signal acquisition and robot motion, there is the need for prediction of the measured motion. In this work, we present a slight modification of the Recursive Least Squares (RLS) algorithm for the prediction of human respiration. We have modified the RLS algorithm as to include an exponential memory term to cope with signal irregularities and system noise. The new algorithm is evaluated on synthetic and real breathing motion signals. The prediction results compare favourably to other methods like Least Mean Squares prediction, Wavelet-based Multiscale Autoregression and Multi-step Linear Methods.