肺CT序列中的含隐变量三维模式病灶识别算法

项目名称： 肺CT序列中的含隐变量三维模式病灶识别算法

项目编号： No.61301257

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 无线电电子学、电信技术

项目作者： 王青竹

作者单位： 东北电力大学

项目金额： 24万元

中文摘要： 肺断层CT序列构成一个三维张量，为了更符合三维CT序列的内部结构以改善检测效果，课题将不局限于二维矩阵理论，而从高维（大于等于三维）张量理论着手，研究一种适合的处理方案。在目标分割方面，拟在高维张量处理规则下建立三维动态轮廓跟踪方法，并通过寻找目标体纹理特征与三维矩阵降秩效果之间的关系来进行优化；在识别方面，针对判断肺部病灶的准确性受制于传统特征所包含的信息有限这一问题，考虑到病灶与肺内其他组织之间的关系对识别及分类的重要辅助作用，拟构建一种训练模式，可以在体现病灶自身三维特征的基础上，隐含其与肺区整体之间的关系。课题的开展有望突破传统高维张量向低维转换所带来的结构破坏和内存浪费等局限，构建可以直接处理三维对象的分割与识别方案，并应用在肺部CT的病灶检测中，为三维医学图像处理领域开拓一种新的途径。

中文关键词： 高维奇异值分解；隐支持张量机；张量处理；计算机辅助诊疗；

英文摘要： Lung CT set constructs a Three Dimensional (3D) tensor. In order to satisfy internal structure of 3D lung CT and achieve better performance, detection scheme based on 3D matrix patterns would be studied by improving higher-order tensor theory rather than traditional 2D matrix theory. On the segmentation, suitable 3D dynamic contour tracking scheme would be built based on higher-order tensor processing rules, and optimized by finding the connection between texture features and reduced-rank effect of the 3D object. On the recognition, improvement aims at the problem that judgment of the lesion is difficult only from traditional features with limited information. As relationship between the lesion and other structures in the lung play important role for identification，suitable training model would be explored adding the latent relationship based on 3D features of the lesion itself. The project is expected to break the limitation of structure destruction and memory wasting caused by the transforming from the higher-order tensor to lower-order model in the traditional scheme. Segmentation and recognition frames based on higher-order tensor theory would be built and used in the lesion detection of lung CT to explore a new approach in the 3D medical image processing.

英文关键词： Higher-order Singular Value Decomposition；Latent STMs；Tensor processing；Computer-aided Diagnosis；