Respiratory Motion Prediction Of Chest And Abdomen Based On CNN And GPR

In the process of radiotherapy,whether the radiation can be accurately irradiated to the tumor area is an important factor that affects the curative effect.For chest and abdomen tumors,because the positions of tissues and organs are constantly changing with respiratory movement,the tumor moves out of the target area or the normal tissue moves into the target area,which brings great difficulties to accurate radiation irradiation.Predicting the breathing movement,and then adjusting the position and angle of radiation exposure,can realize dynamic tracking radiotherapy,which greatly improves the effect of chest and abdomen radiotherapy.Existing respiratory motion prediction models mainly include two categories:model algorithms and model-free algorithms.Among them,the premise assumptions when building a model based on the model algorithm are not necessarily correct,and the set constants also need to be updated at any time.It is difficult to achieve real-time prediction of respiratory motion during radiotherapy;As for sample individuals with different respiratory states,the algorithm of no model-free respiratory motion prediction can realize adjustment of parameters automatically,update different test samples quickly.It can also adapt to respiratory movements with non-strict periodic characteristics,but the prediction accuracy needs to be improved.Respiratory motion is non-strictly periodic,and Convolutional Neural Networks(CNN)has a good predictive effect on repetitive data.If the CNN model is selected to extract the local features of the respiratory motion signal and integrate it into the overall feature.Using the correlation of the acquired data to predict respiratory movement will help improve the prediction accuracy of the respiratory movement.To this end,this paper establishes a respiratory motion prediction model based on CNN and Gaussian process regression(GPR).Firstly,preprocess the collected chest and abdomen respiratory motion signals to meet the input data requirements of the CNN prediction model;then image the preprocessed data,using the feature extraction function of CNN,and good fault tolerance and parallel processing ability to realize the respiratory motion prediction model;finally,the GPR algorithm is combined to adjust the prediction value and offset the over-fitting problem of CNN,thereby further improving the prediction accuracy of the model.Through the comparison between experimental example and the model without GPR algorithm error correction,KNN model,BP-NN model,AOSVR model of the respiratory motion prediction results,using the mean absolute percentage error(MAPE)and root mean squared error(RMSE)to evaluate the prediction effect of the model.The average value of MAPE 0.2928 and average RMSE 0.0345 of the prediction results of the method in this paper are smaller than the prediction model without GPR algorithm,KNN model,BP-NN model,and AOSVR model.It shows that the method in this paper improves the accuracy of respiratory motion prediction.