Research On External And Internal Correlation Of Respiratory Movement Based On Multi-task Gaussian Process

Radiotherapy is one of the important methods in oncology.It is hoped that while the therapeutic beam fully irradiates the tumor target area,it is necessary to avoid irradiating normal cells around the tumor as much as possible.However,the influence of a series of physiological factors,mainly respiratory movement,cause the tumor target of the chest and abdomen movement,so the tumor must be tracked in real time during the treatment.Respiratory tracking based on external signal uses the correlation between external and internal signals to establish an association model,and indirectly predicts the movement of the tumor through external signals,thereby reducing the damage of radiotherapy to patients.It is a widely used tumor tracking method,and the focus is on the predictive ability of the association model.This thesis first studies the internal and external respiratory motion signals.According to the modeling requirements,the data is intercepted,denoised and normalized.Comprehensively analyze the correlation between the internal and external motions of different individuals,different positions and different respiratory stages,so as to provide a data basis for the establishment of the correlation model.Then,we analyze the working process of the tumor tracking system.According to the role of the internal and external correlation model in the system,the structure of the model and the establishment process are designed.We determine the time series form of input and output data and regarding the specific research samples,establish the correlation model by using dual-polynomial method which is widely used in this field.Finally,the correlation model based on the Multi-task Gaussian Process is established according to the characteristics of the data as multi-external motion signals.From the perspective of the comparison of model-based and model-free methods,single-task and multi-task methods,using the mean absolute error and root mean square error as evaluation indicators,the prediction results are compared with that of dual-polynomial method and Gaussian Process Regression to verify the predictive ability of the correlation model.