Research On Three-Dimensional Visual Imaging Of Thoracic And Abdominal Surface And Machine Learning Prediction Of Respiratory Motion

Radiotherapy,as one of the three main treatment methods,plays a vital role in cancer treatment.In recent years,advances in radiotherapy technology are closely related to accurate tumor location.It is extremely important i n clinical radiotherapy to ensure that the tumor target area is fully irradiated and the irradiation of normal tissues and organs around the tumor is reduced as much as possible.However,respiratory movement happens during radiotherapy,the displacement and deformation of tumor would occur,especially in the chest and abdomen.Therefore,it is paramount to improve the effect of radiot herapy by tracking the respiratory movement of thoracic and abdominal tumors and implementing the supplement and management of respiratory movement in clinical radiotherapy.Medical imaging has shortcomings such as poor real-time in vivo anatomical imaging such as tumors,patients with additional doses of radiation,and invasive implantation of landmarks in the body.So it is not suitable to apply direct tracking respiratory movement in clinical radiotherapy stage.Therefore,it is recognized as the most practical and promising development to realize respiratory motion compensation and management in radiotherapy.At present,the research of tracking in-vitro respiratory movement focuses on the measurement and prediction of landmark respiratory movement.Howe ver,two problems occur.One is the difficulty in describing the respiratory movement on the surface of chest and abdomen comprehensively through a single point or several points,and to predict the respiratory movement accurately is even more unachievable.The other is that the existing prediction algorithms are difficult to adapt to the characteristics of severe nonlinear and few sample points of chest and abdomen breathing movement,and to meet the requirements of both prediction accuracy and real-time in clinical radiotherapy is even harder.To solve the two problems,this thesis studies the 3D vision imaging method of thoracoabdominal surface based on fringe phase and the breathing motion prediction algorithm of thoracoabdominal surface based on relevan ce vector machine.The basic measurement model of 3D imaging system is established to provide theoretical foundation and mathematical means.In order to improve the accuracy of wrap phase extraction,a local fringe signal wrap phase measurement model based on improved Morlet wavelet is established.A package phase unwrapping model is established,which provides periodic error s uppression function in principle.To achieve high accuracy imaging,a 3D imaging method based on the fringe phase of asymmetric digital code and analog code combination is proposed.The phase is used to replace the intensity for digital encoding and decoding to improve the anti-interference ability.The asymmetric combination between digital code bits and between digital code and analo g code is used to improve the periodic error suppression ability.An asymmetric digital coding and decoding model based on phase and an asymmetric digital analog code combination coding and decoding model is constructed,and its period error suppression ability is analyzed.In terms of the prediction algorithm of breathing movement on chest and abdomen surface aiming at small sample size,accuracy and real-time performance,a prediction algorithm of body surface breathing movement based on relevance vector machine and multitask Gaussian process is proposed.The Gaussian process is introduced into the chest and abdomen surface br eathing movement prediction,which is suitable for small samples and takes the accuracy and real-time into account.The single task is extended to multitask to improve the prediction accuracy by using the three-dimensional coordinates of the landmark space.The basic principle of the algorithm is described,and the specific implementation scheme of the algorithm is given.In order to h ighlight the nonlinear adaptability,a correlation model of internal and external respiratory movement based on multi kernel function relevance vector machine is proposed.The correlation analysis of internal and external respiratory movement signals is carried out.According to its characteristics,multi kernel function is introduced into relevance vector machine to improve the nonlinear fitting ability of correlation model and the ability to capture model features in small samples.In order to improve the prediction ability of the correlation model,the fruit fly optimization algorithm is used to optimize the model structure,and an in-vivo and in-vitro respiratory motion correlation algorithm is formed.The evaluation index and modeling principle of the correlation model are given.Directed at the above methods and theoretical research results,numerical simulation,simulation measurement and actual measurement are used to verify the experimental results.Numerical simulation and actual measurement verify the correctness of the basic measurement model of the 3D imaging system based on fringe analysis.Numerical simulation shows the effectiveness and superiority of the proposed step phase code.The measurement of human thorax and abdomen surface verifies the effectiveness of the three-dimensional imaging method of human thorax and abdomen surface breathing motion based on wavelet.Simulation measurement and human chest and abdomen surface measurement quantitatively verify the effectiveness and superiority of the asymmetric step phase code and phase shift code combined 3D imaging method based on fringe phase,and plaster head image measurement verifies its ability of3 D imaging of complex surface.The experimental results verify the effectiveness of the surface breathing motion prediction algorithm based on the combination of relevance vector machine and multi-task Gaussian process.The model and its nonlinear fitting ability in small samples are verified.This study will provide a novel method for full field three-dimensional imaging of thoracoabdominal surface.A theoretical basis and technical means will provide for description and prediction of breathing movement of thoracoabdominal surface based on many points and regional features.A new algorithm and model will provide for prediction of in-vivo and in-vitro breathing movement of thoracoabdominal.A theoretical basis and technical means will provide for compensation and management of breathing movement in radiotherapy.They significantly improve the effect of radiotherapy and the health level of the people.