Research On Prediction Of Thoracoabdominal Surface Respiratory Motion Based On GRA And GPR

During radiotherapy of thoracic and abdominal tumor,respiratory movement will make the tumor leave the target area or normal tissue into the target area,resulting in incomplete irradiation of tumor or radiation damage to normal tissue,seriously affecting the effect of radiotherapy.At present,dynamic tracking technology can solve the above problems.In order to achieve real-time tracking,accurate prediction of tumor location is critical.The early model prediction method is difficult to adapt to the irregularity of respiratory motion,and the error is large.Therefore,scholars began to study the non-parametric prediction method based on learning,and achieved certain results,but the prediction accuracy still needs to be improved.In order to further improve the prediction accuracy,this paper proposes a prediction method of thoracoabdominal surface respiratory motion based on grey relational analysis(GRA)and Gaussian process regression(GPR).The main research contents are as follows :Firstly,the respiratory motion data is collected and preprocessed.The respiratory motion data acquisition system was built based on FASTRAK.Considering that the respiratory motion data can not be collected by radiotherapy marker points during radiotherapy,an experimental scheme for predicting radiotherapy marker points with auxiliary marker points is developed.The respiratory motion data were collected and processed by deleting abnormal values,smoothing denoising and normalization.Then,the GRA model of auxiliary marker points and radiotherapy marker points is established.The respiratory motion data of the radiotherapy marker points were used as the comparison sequence,and the X-axis,Y-axis,and Z-axis data of the auxiliary marker points were used as the reference sequences,respectively.Three gray correlation analyses were performed to obtain the correlation degree of each coordinate axis of the radiotherapy marker point relative to the X-axis,Y-axis,and Zaxis data of the auxiliary marker points.The mapping relationship between the two points was established using the weight of the correlation degree.Third,build a GPR prediction model.The auxiliary marker point data is divided into training set and test set.The square exponential covariance function is selected as the kernel function to initialize the hyperparameters,and the GPR prior model is trained.Then the conjugate gradient method is used to solve the optimal hyperparameters,and finally the GPR posterior model is determined.Finally,the experimental verification of the method is completed.Based on 200 sets of data,the GPR model was used to predict the respiratory motion of the auxiliary marker points,and then the GRA model was used to map the respiratory motion from the auxiliary marker points to the radiotherapy marker points.The mean absolute percentage error(MAPE)and root mean square error(RMSE)were used as evaluation indexes to compare the proposed method with linear prediction,BP neural network prediction and support vector machine regression prediction.The results are as follows : the average MAPE of the proposed method is 0.3056,which is smaller than 1.9607,1.1421 and 1.6338 of the other three methods.The average value of RMSE obtained by this method is 0.0334,which is less than 0.1336,0.0776 and 0.1140 of other methods,indicating that the prediction accuracy of this method has been improved.