Research On Segmentation Of Organs At Risk On Head And Neck Of CT Images Based On Deep Full Convolution Neural Network

At present,radiation therapy has been widely used in the treatment of head and neck tumors.Doctors need to segment protected organs at risks(OARs)on CT to reduce radiation damage to healthy tissue.Now some methods of semi-automatic or fully automatic segmentation have appeared,but still have some limitations.But deep fully convolutional neural network(FCN)provides a new solution for the automatic segmentation of OARs in head and neck(H&N)CT and has practical significance.This thesis is based on the demand of radiotherapy intelligent segmentation system of Shenzhen Yino Intelligent Technology Co.Ltd.Research the segmentation method of H&N OARs based on FCN.The automatic segmentation and management subsystem is realized for providing a reliable automatic segmentation tool for doctors.This thesis includes the following main aspects:1.proposing an optimization method of segmentation of H&N OARs based on2D FCN: Introducing batch normalization layer to solve inconsistent datadistribution.Using channel attention mechanism to strengthen importantfeatures.Proposing a new loss function to solve the class imbalanceproblem.2.proposing a method of segmentation of H&N OARs based on 2D and 3Dcombined FCN: Proposing a new network structure to combines 2D and 3Doperations for learning edge and semantic features.Proposing a new methodto improve the segmentation of small organs.3.developing a H&N OARs automatic segmentation service and managementsubsystem: Providing a convenient automatic segmentation tool for doctors.Supporting functions such as management of segmentation models andmonitoring of segmentation task state.