Research On Clinical Target Delineation Method For Rectal Cancer Radiotherapy Baesd On Tissue Prior

Rectal cancer radiotherapy clinical target volume delineation is a key step in clinical radiotherapy.In actual clinical practice,doctors usually perform manual delineation according to the clinical target volume delineation guidelines for rectal cancer radiotherapy.Manual delineation takes a long time,which greatly increases the workload of doctors.With the development of computer technology and the maturity of deep learning,automated clinical target volume delineation of rectal cancer radiotherapy has become a popular research direction.Therefore,combining rectal cancer radiotherapy clinical target delineation guidelines with deep learning to improve the accuracy of automated radiotherapy clinical target volume delineation is the focus of this thesis.This thesis takes the abdominal CT images of patients with rectal cancer as the research object,and constructs a deep learning network framework that combines contextual information,a dual attention module based on image classification and surrounding tissue boundaries,and a topological structure learning based on the contours of the clinical target volume.First of all,this thesis slices the three-dimensional image horizontally,and takes two-dimensional images as the research object of this thesis.The features of voxel on the two-dimensional image is not only related to the features of adjacent voxels on the current image,but also related to the features of near voxels on the upper and lower two-dimensional images.In order to fuse the information between different layers of twodimensional images’ contexts,this thesis uses two-dimensional convolution to learn the features of each layer of the two-dimensional image,and uses one-dimensional convolution to learn the interlayer features of the multi-layer two-dimensional images.Second,based on the fact that not all twodimensional images have the clinical target volume of rectal cancer,the two-dimensional images can be divided into images containing the clinical target volume and images not containing the clinical target volume.And according to the guidelines for delineating the clinical target volume of rectal cancer,the boundary of the clinical target volume is defined based on the boundary of the surrounding tissue.Based on the above two features,in order to help the network to focus on learning the features of the two-dimensional image containing the clinical target volume,it focuses on learning the features of the voxels on the boundary of organ at risk that are closely related to the boundary of the clinical target volume.A dual attention module based on image classification and surrounding tissue boundaries is used in network.Finally,the traditional convolution operation uses the same calculation feature method for all voxels on the image.In order to further learn the features of the nodes on the boundary of the clinical target volume and use the boundary topology,this thesis adopts the clinical target volume boundary topology and features learning module based on graph convolution.The boundary topology and feature learning module is used to perform features learning on the nodes on the boundary of the multi-layer clinical target volume,and learn the contour of the clinical target volume in the layer and the change trend of the clinical target volume boundary between the layers.In order to evaluate the performance of the tissue priori-based rectal cancer radiotherapy clinical target volume delineation method constructed in this thesis,this thesis conducted experiments on the CT image data set of 64 patients with rectal cancer provided by Shandong Cancer Hospital.A qualitative comparison and quantitative analysis of the experimental results show that the method constructed in this thesis can achieve better clinical target volume delineation results for rectal cancer radiotherapy than other methods.