Spatiotemporal Constraint Prediction Of Respiratory Motion In Human Thoracic And Abdominal Surface Areas During Radiotherapy

During the radiotherapy of thoracoabdominal tumors,breathing movement will change the volume and position of the tumor target area,causing the tumor to overflow the target area or normal tissue to enter the target area.Carrying out research on the prediction of respiratory movements is conducive to the dynamic tracking of radiotherapy for thoracoabdominal tumors and improves the effect of radiotherapy.Early prediction methods considered breathing movement as a simple repetitive single-cycle movement,but because breathing does not have a strict periodicity,the prediction results deviate greatly from the actual situation;later,a model prediction method appeared,based on breathing movement Historical data,through the establishment of mathematical models,to realize the prediction of future respiratory movements,this method can overcome the non-strict cyclicity of breathing and effectively improve the prediction accuracy;in recent years,non-parametric prediction methods have emerged,which can make full use of historical data The coupling relationship between them further improves the prediction accuracy.Among them,the Gaussian process regression prediction method provides the statistical results of respiratory prediction in the form of mean and variance,which provides a new technical method for respiratory motion prediction.However,Gaussian process regression itself is a mathematical method.If it is combined with the characteristic parameters of respiratory motion,it will be helpful to further improve the prediction accuracy.In this paper,the temporal and spatial thresholds of respiratory motion are used as constraints,and a Gaussian process regression breathing motion prediction method based on spatial and temporal constraints is proposed.By using the correlation information of respiratory motion time and space,and using the temporal and spatial range of respiratory motion,The research purpose of improving prediction effect was achieved.The main research contents of the paper are as follows:1.Respiratory data acquisition and constraint parameter acquisition.FASTRAK motion tracking and positioning system was first used to collect respiratory signals,and then the principle of temporal and spatial constraints to improve prediction accuracy was expounded according to the characteristics of respiratory motion.Since the temporal and spatial distributions were limited to reasonable in the process of respiratory motion prediction Within the threshold range,it can filter out the unreasonable results contained in the traditional Gaussian process regression prediction method,and then improve the prediction accuracy.Finally,based on the breathing signal,obtain the experimenter's time and space constraint parameters.2.Establish a Gaussian process regression model for respiratory motion prediction based on space-time constraints.First,the basic principles of Gaussian process regression are introduced.Then,from the perspective of weight space,the prediction range of the traditional Gaussian process regression model is constrained in time and space,and the time constraint parameters and space constraint parameters are combined with the Gaussian process regression method,respectively.,Construct a Gaussian process regression prediction model based on time constraints and a Gaussian process regression model based on space constraints;finally,combine the above two models to meet the simultaneous constraints in time and space,construct a Gaussian process regression prediction based on space-time constraints model.3.Prediction experiments and comparative analysis.The prediction results of the Gaussian process regression prediction model based on spatio-temporal constraints are compared with the prediction results of the traditional Gaussian process regression model without constraints and the prediction results of non-parametric regression,linear prediction and BP neural network prediction algorithms.The experimental results show that the relative error range of the prediction results of this method is 0.08%,which is less than 0.31%,0.46%,0.46%,and 0.55% of the traditional Gaussian process regression model,BP neural network,linear prediction and non-parametric regression.The stability is better than the above four comparison methods;the mean value of the root mean square error(ARMSE)is used to evaluate the error between the prediction result and the real respiratory motion.The ARMSE of this method is 0.1547,which is lower than the traditional Gaussian process regression,nonparametric regression,The linear prediction and BP neural network of 0.2145,0.894,0.5651,0.5207 show that the prediction accuracy of the method in this paper is higher than the above four comparison methods.