面向多时相腹部CT图像的多器官计算机辅助诊断关键技术研究

The simultaneous automatic segmentation and comprehensive analysis of liver, spleen and kidneys from abdominal CT series are the important prerequisite for abdominal diseases diagnosis, pre-operative planning and guidance. Aiming at the inaccurate automatic segmentation of multiple organs because of the insufficient information from single phase CT series, the project proposes a multiphase-based 4D graph cuts algorithm for the automatic segmentation of abdominal multiple organs from multiphase CT series. Firstly, combining the anatomy prior models and the intrinsic information of multiphase CT series, this project constructs the energy function of 4D graph cuts, and then based on the brightness characteristics of target organs from the multiphase abdominal CT series, locates the seed regions of the target organs, and realizes the simultaneous robust automatic segmentation of multiple organs from abdominal CT series based on 4D graph cuts. Next, this project constructs the feature vector space for the liver series obtained from the 4D graph cuts algorithm by combining the 2D texture information and the 3D voxel information of the liver series, and extracts liver blood vessels from the segmented liver series using back-propagation neural network and level set evolution method, and constructs 3D liver vascular visualization system, where the hepatic artery, the hepatic portal vein and the hepatic vein are distinguished, respectively. This project aims at the the key technologies for abdominal multiple organs of volume measurement, 3D reconstruction, diseases diagnosis, surgical navigation, organ transplantation, and liver vascular recognition and analysis, which help the radiologists and clinicians timely and easily get the overall information and 3D visualization of abdominal multiple organs (tissues) simultaneously, and provide the technical supports and decision-making services for computer-aided diagnosis and diseases treatment.

对腹部CT序列图像肝脏、脾脏以及肾脏的同步自动分割和综合分析是腹部疾病诊断、术前规划与指导的重要前提。针对单时相CT序列因信息量不足使得多器官同时自动分割不准确的问题，提出基于多时相的4D图割算法。首先结合解剖学先验模型和多时相CT图像本征信息，构建4D图割能量函数，然后根据多时相CT序列目标器官亮度特征，定位种子区域，实现基于4D图割的多器官同时鲁棒自动分割。根据肝脏分割结果，结合肝序列图像的二维纹理信息和三维体数据信息建立特征空间，运用反向传播神经网络和水平集演化方法对已分割出的肝脏进行血管提取，建立血管三维显示系统，并对肝动脉、门静脉和肝静脉进行识别。本项目研究可解决腹部多器官的体积测量、三维重建、疾病诊断、手术导航、器官移植以及肝脉管识别与分析等所需的关键技术问题，有助于医师及时方便地同时获取腹部多器官(组织)的整体信息和三维显示，为计算机辅助诊断和疾病治疗提供技术支持和决策服务。

对腹部CT序列图像肝脏、脾脏以及肾脏的同步自动分割和综合分析是腹部疾病诊断、术前规划与指导的重要前提。针对单时相CT序列因信息量不足使得多器官同时自动分割不准确的问题，提出了几种可行的腹部器官自动分割方法。主要包括肝脏(含病态肝脏)的自动分割，肝脏血管的自动分割与识别，肾脏和脾脏的自动分割等。重点研究了肝脏的自动分割以及基于此的肝脏血管系统的分割、三维重建、肝静脉与肝门静脉的辨识等。.本项目的主要研究内容与成果包括：1)提出了基于图割和瓶颈检测的腹部CT序列图像肝脏自动分割方法，以及基于边缘行进的肝脏血管补偿方法；2)提出了基于极限学习机的肝脏血管分割方法；3)提出了基于OOF和OFA的肝静脉和肝门静脉分割与辨识方法，以及基于中心线约束与强度模型的肝脏血管分割方法；4)提出了一种基于bi-Gaussian滤波器和活动轮廓模型的肝脏血管自动分割方法；5)提出了基于空间模糊C均值与区域生长的腹部CT序列图像肾脏自动分割方法；6)提出了基于卷积神经网络的腹部CT序列图像多器官自动分割与识别方法；7)提出了一种腹部CT序列图像多器官的半自动分割方法。.本项目研究可解决腹部多器官的体积测量、三维重建、疾病诊断、手术导航、器官移植以及肝脉管识别与分析等所需的关键技术问题，有助于医师及时方便地同时获取腹部多器官(组织)的整体信息和三维显示，为计算机辅助诊断和疾病治疗提供技术支持和决策服务。.通过本项目的研究，已发表相关学术论文18篇，其中SCI检索论文11篇，EI检索论文6篇；还有2篇国际期刊SCI论文已进入二审；已授权国家发明专利1项，软件著作权1项，已申请国家发明专利1项；获中南大学优秀博士学位论文1篇，湖南省自然科学二等奖1项。