Motion prediction of lung tumor using Predicted Error-based Normalized Least Mean Square algorithm

In robotic radiotherapy, one of the problems is systematic latencies between the acquisition of the target position and the mechanical response of the system to follow the target position. To compensate the aforementioned latencies while tracing the tumor motion, an accurate algorithm to predict these latencies is therefore required. The prediction algorithm computes the future target position. In this study, we have analyzed the accuracy of three algorithms that predict tumor positions with sufficient lead time to compensate these systematic latencies. The motions have been analyzed for predictability up to 400ms in advance using Least Mean Square (LMS) prediction, Normalized Least Mean Square (NLMS) prediction and the proposed algorithm, named as a Predicted Error-based Normalized Least Mean Square (PE-NLMS) prediction. The performance of three prediction algorithms is evaluated using three real breathing signal data in 30 Hz sampling rate. The results show that the PE-NLMS is outperformed with respect to RMSE by all other algorithms and does not require too much parameter adjustment. The simulation showed that it Jitter needs to be improved for better performance.