A Training Strategy for Enhancing the Accuracy of Real-Time Tumor Tracking Based on Deep Bi-LSTM Learning

Radiotherapy of thoracic-abdominal tumors requires high accuracy to avoid radiating normal tissues, which will lead to a series of complications. The accuracy of the prediction of respiratory motion can improve the accuracy of tumor location. The Deep Bi-LSTM-based breathing curve prediction method has a better effect than the traditional prediction method. This paper aims to discuss the impact of data volume improvement on respiratory prediction accuracy. Moreover, the respiration motion for a latency about 380ms is predicted by a bidirectional long short term memory (Deep BiLSTM) model. In this study, 98 malignant lung tumor motion of patients is acquired to train and verify the accuracy of the model. It is demonstrated that the LSTM-based breathing curve prediction method has a better effect than the traditional prediction method. Furthermore, comparative experiments are used to discuss the influence of the increase in the amount of model training data on the accuracy of the model.