A Comparative Study of a Novel AE-nLMS Filter and Two Traditional Filters in Predicting Respiration Induced Motion of the Tumor

摘要

Prediction of tumor motion is one of the important steps in active tracking of tumor and dynamic delivery of radiation dose to tumor. In this paper, we have presented a novel adaptive acceleration-enhanced normalized least mean squares (AE-nLMS) prediction filter based on the adaptive normalized least mean squares (nLMS) filter with predicted acceleration and ratio between the real and predicted acceleration taken into account. We have compared the performances of nLMS, artificial neural network (ANN), and AE-nLMS filter for predicting the respiration motion during normal and irregular respiration. The results revealed that the ANN filter has the best performance in the prediction of normal respiration motion, whereas the AE-nLMS filter outperformed other filters in the prediction of irregular respiration motion.