Research On Respiratory Motion Prediction Model For Target Position Tracking In Radiotherapy

Tumor is one of the main diseases affecting human health,and the radiotherapy system is the main means of treating tumors.In the radiotherapy of thoracic and abdominal tumors such as lung cancer and gastric cancer,the breathing movement of the human body will cause the position of the tumor to change,which in turn affects the accuracy of the position of the radiotherapy target area.make an impact.Aiming at this problem,this paper proposes a multi-scale enhanced time series convolution respiratory motion prediction model based on empirical mode decomposition,which can improve the prediction accuracy of the positioning system formed based on in vitro markers,and solve the problem of system delay on the prediction accuracy.Influence.In this paper,the empirical mode decomposition method is used to reduce the influence of non-stationary and nonlinear respiratory motion signal on the prediction model.By studying the performance of different empirical mode decomposition methods and the effect of improving the prediction ability of the model,it is verified that the empirical mode decomposition method can effectively solve the problem of simplification of respiratory motion signal features,and can improve the accuracy of respiratory motion prediction models.According to the feature components formed after empirical mode decomposition,a respiratory motion prediction model based on the fusion of multi-scale convolutional neural network and deep time-series convolutional network based on attention mechanism is proposed.The model uses multi-scale convolutional layers to extract features in parallel,finds the optimal local sparse structure of the convolutional network,and fully obtains time series information.At the same time,the attention mechanism is integrated into the convolution channel features to improve the model’s sensitivity to the convolution channel features,and independently learn different weight coefficients for each channel feature.By introducing a time-series convolution mechanism to capture the spatial dependencies of respiratory motion data features,the efficiency of respiratory motion prediction models is improved.The model in this paper is compared with various deep prediction models such as CNN-Bi＿LSTM,CNNs-TCN,and EMD-CNNs-Bi＿LSTM in terms of prediction accuracy and time efficiency.The results show that under the condition of a delay time of 450 ms,the mean absolute error and root mean square error of the prediction model are reduced by about 12% on average,the overall fit is improved by about 1.5%,and the network update time is shortened by about 60%.The model proposed in this paper has superior prediction performance under different input data length and delay time.