Prediction Of Respiratory Movement On Thoracoabdominal Surface Based On CNN-LSTM

In the process of radiotherapy for tumors in the chest and abdomen,the organs and tissues in the chest and abdomen will constantly change position with respiratory movement.Finally,the respiratory motion leads to the tumor moving out of the radiation target area or the healthy tissue entering the radiation target area,which affects the accuracy of radiotherapy.Predicting respiratory motion and obtaining real-time location information of tumor is an effective means to realize dynamic tracking radiotherapy and improve the effect of radiotherapy.At present,the research on respiratory motion prediction mainly includes modelbased prediction algorithms and model-free prediction algorithms.The model algorithm needs to make assumptions and evaluate the model in advance.Because the parameters of the model are fixed and can not be adjusted according to the change of respiratory motion,it leads to a large error between the predicted results and the actual respiratory motion.The model-free algorithm does not need to know the change law of respiratory motion in advance,and can automatically adjust the parameters according to the respiratory motion,and the accuracy is improved.In order to further improve the prediction accuracy,this paper proposes a prediction method of respiratory movement on the chest and abdomen surface based on CNN-LSTM.The main research contents are as follows:Respiratory signal acquisition and preprocessing.Firstly,FASTRAK is selected to build a respiratory movement data acquisition device on the chest and abdomen surface,and 300 groups of respiratory movement data are collected.Then,the data are preprocessed,the box plot method is used to remove the impulse noise in the data,and the S-G smoothing filter is used to smooth and de-noise the data,and the processed data set is divided.CNN and LSTM were combined to build a CNN-LSTM respiratory motion prediction model.Firstly,the basic principles of CNN network and LSTM network were introduced respectively.Then,the CNN-LSTM respiratory prediction model is built,the preprocessed respiratory motion data is cut into fixed size samples,and the spatial structure is adjusted as the input of the CNN layer.The CNN network is used to extract the correlation of the three-dimensional coordinates of the respiratory signal and the timing features of the respiratory signal,including the global features of the respiratory motion signal.The LSTM network was used to model and classify the feature information,and the deeper time series features in the respiratory motion signal were mined to predict the respiratory motion in real time and improve the prediction accuracy of respiratory motion.Finally,the parameters of each layer of the CNN-LSTM prediction model were initialized.The CNN-LSTM respiratory motion prediction model was trained.Firstly,the model training strategy was formulated,including configuring the development environment and formulating the training process.Then,the model parameters were selected,the Re LU function was selected as the activation function,the mean square error was selected as the loss function,Adam was selected as the parameter optimizer,and the Dropout method was selected for regularization.Finally,the model parameters were adjusted and determined based on the network search method.Complete the prediction experiment.Firstly,root mean square error(RMSE)and mean absolute percentage Error(MAPE)were determined as the evaluation indexes of respiratory movement prediction.Then,the prediction results of the method based on the paper are compared with the LSTM method without CNN,BP neural network,support vector regression and Gaussian process regression.The comparison results are as follows:The average RMSE of the prediction results of the method in this paper is 0.0218 mm,and the average MAPE is 0.2052 mm,which are smaller than the other four prediction algorithm models,proving that the prediction model based on CNN-LSTM proposed in this paper can improve the prediction accuracy of respiratory movement.