原发性肝癌多模态图像辅助诊疗自动化分析的关键技术研究

项目名称： 原发性肝癌多模态图像辅助诊疗自动化分析的关键技术研究

项目编号： No.U1401254

项目类型： 联合基金项目

立项/批准年度： 2015

项目学科： 管理科学

项目作者： 方驰华

作者单位： 南方医科大学

项目金额： 242万元

中文摘要： 原发性肝癌死亡率居广东省恶性肿瘤首位，外科手术是首选。CT或MR单一模态成像都不能完整提供精确的解剖信息，三维重建、手术规划的准确性和可靠性低，难以指导精准手术。基于此，课题组在肝胆胰肿瘤三维重建基础上进行研究：（1）构建多模态图像大数据和先验知识模型；（2）利用变分法分割脉管，根据解剖学原理和门静脉分型改进肝脏分段；（3）利用图谱、概率等先验信息约束，在非线性配准中整合统计形状模型，自动分割肝脏；（4）基于肝脏运动各向异性原理正则化多期CT图像配准，联合几何和灰度特征进行CT-MR非线性配准；（5）构建多模态图像局部纹理、上下文、图谱相关特征，优化稀疏与低秩表示，提取肝脏肿瘤；（6）制订与临床需求适应的评价和验证准则，开放数据。在非线性配准框架下，整合肝脏形状、呼吸运动、血管特征等进行正则化约束，使配准更符合物理规律，实现精准图像信息融合、分割，为肝脏外科提供精准的数字化信息和技术支持。

中文关键词： 图像分割；图像配准；机器学习；肝段划分；开放数据

英文摘要： The mortality rate of hepatocellular carcinoma ranks first in Guangdong province, surgery is the preferred treatment method. Mono-modality CT or MR image could not provide complete and accurate anatomical information. Three-dimensional (3D) reconstruction and surgical planning have low accuracy and reliability so that they are difficult to guide surgery precisely. To solve the above problems, research team will conduct the following studies based on preliminary 3D reconstruction of hepatobiliary and pancreatic tumors: (1) constructing of big multimodality image data and a prior knowledge model; (2) extracting vessel by the principle of calculus variation and improving hepatic segment division according to principles of anatomy and portal vein subtype; (3) integrating of statistical shape model into nonlinear registration with atlas and a prior probability constraints for automatic liver segmentation; (4) regularizing anisotropic liver respiratory motion principle into nonlinear registration of multiphase CT images; jointing geometrical and grayscale image features into CT and MR nonlinear registration; (5) extracting liver tumor via optimization of low rank and sparse representation from multimodality local texture, context, atlas-related features; (6) Formulating criteria of evaluation and validation for image analysis to satisfy clinical requirement, opening and sharing data. Integration regularization constraints of liver shape, respiratory motion, vascular characteristics and other features into nonlinear registration framework to make registration more in line with the laws of physics to achieve precise image fusion and segmentation and to provide accurate digital information and technical support for liver surgery.

英文关键词： Image segmentation;image registration;machine learning;liver segment division;open data