Intelligent sensing and monitoring : respiratory motion prediction for tumor following radiotherapy

This paper presents a medical application of the intelligent sensing and monitoring, a new lung tumor motion prediction method for tumor following radiation therapy. An essential core of the method is accurate estimation of complex fluctuation of time-varying periodical nature of lung tumor motion. Such estimation is achieved by using a novel multiple time-varying seasonal autoregressive (TVSAR) model in which several windows of different time-lengths are used to calculate correlation based fluctuation of periodic nature in the motion. The proposed method provides the prediction as a combination of those based on different window lengths. Multiple regression (MR), multilayer perceptron (MLP) and support vector regression (SVR) are used to combine and the prediction performances are evaluated by using clinical lung tumor motion. The proposed methods with the combined predictions showed high accurate prediction and are superior to the single different predictions. The average errors of MR, MLP, and SVR were 0.8455,0.8507, and 0.7530 mm at 0.5 s ahead, respectively. The results are clinically sufficient and thus clearly demonstrate that the proposed TVSAR with an appropriate combination method is useful for improving the prediction performance.