Prediction of respiratory motion using wavelet based support vector regression

In order to successfully ablate moving tumors in robotic radiosurgery, it is necessary to compensate the motion of inner organs caused by respiration. This can be achieved by tracking the body surface and correlating the external movement with the tumor position as it is implemented in CyberKnife^® Synchrony. Due to time delays, errors occur which can be reduced by time series prediction. A new prediction algorithm is presented, which combines á trous wavelet decomposition and support vector regression (wSVR). The algorithm was tested and optimized by grid search on simulated as well as on real patient data set. For these real data, wSVR outperformed a wavelet based least mean square (wLMS) algorithm by >; 13% and standard Support Vector regression (SVR) by >; 7:5%. Using approximate estimates for the optimal parameters wSVR was evaluated on a data set of 20 patients. The overall results suggest that the new approach combines beneficial characteristics in a promising way for accurate motion prediction.