Prediction Of Respiratory Motion Based On Gaussian Process

One of the important methods to cure cancer is radiation therapy, and the demand of modern radiation therapy is as much as possible to the tumor cancer cells and as much as impossible to avoid tumor surrounding normal cells at the same time. However the influence of uncertain factors in the patients' own breathing and heart beating, making tumor target certain irregular movement. In the course of treatment, the whole system exists a definite delay, because of the data acquisition, transmission, treatment instrument adjustment therapy and other reasons. So real-time tracking and the prediction of tumor target becomes more and more important. To solve the problem, this paper mainly studies the respiratory motion prediction techniques for the main effects of tumor.Because of the irregular and uncertain respiratory motion, a reasonable mathematical model is needed to predict the result. For this purpose, we statistic and analyze the respiratory motion data, and then put forward a respiratory motion prediction algorithm based on the Gaussian process regression. The article research from the following several aspects.First of all, we analyze the human respiratory motion feature, and divide the respiratory signal into three states by using the finite state model. At the same time we use NDI optical instruments to collect the respiratory motion data. After that we pretreat the data and use the data as an input value of the prediction algorithm.Then a model based on Gaussian process model is proposed to predict the regression problem. We use the data to construct a suitable regression model, train and predict. During training, we select the kernel function and find out the super parameter. The feasibility of the algorithm is verified by an off-line simulation experiment.Finally we choice periodic kernel function and work out super parameter by conjugate gradient method. Then the model of respiratory motion data was predicted by the Gaussian Process Regression. After that, the prediction results are compared with three commonly algorithm of the prediction, and the RMSE and the MAPE are used for evaluating result.