脑神经胶质瘤的定量磁共振监测和评价基础研究

项目名称： 脑神经胶质瘤的定量磁共振监测和评价基础研究

项目编号： No.81000635

项目类型： 青年科学基金项目

立项/批准年度： 2011

项目学科： 金属学与金属工艺

项目作者： 喻罡

作者单位： 西安交通大学

项目金额： 10万元

中文摘要： 恶性肿瘤严重威胁人类生命，正确评估其生理功能和状态，对于正确认识肿瘤的行为，引导肿瘤的科学研究和临床治疗具有重要意义。本项目利用定量磁共振成像方法对脑神经胶质瘤进行监测，利用灌注、波谱和DTI成像，获取血流动力学、代谢物水平、扩散水平等参数，构建体现胶质瘤形态和功能的特征。利用基于支持向量机的数据挖掘技术对特征进行优化选择，得到具有较高敏感性和特异性的特征集合，实现胶质瘤分化程度的影像学评价。 本项目首次提出了多模态磁共振定量评价胶质瘤的自动方法，这种方法建立在机器学习和模式分类数学理论上。主要贡献包括：首先，提出一种新的、自动化的、标准化的选取肿瘤感兴趣区域的方法，降低结果对研究者的人为依赖，可以获取稳定可靠的分类器输入数据。第二、利用模式分类理论完成了特征筛选，建立了评价胶质瘤分化程度的分类器，经过测试具有较高的准确性。第三、实现了磁共振数据处理、特征优化选择和支持向量机分类的软件工具，为后续研究和应用提供了支持。本项目证明了多模态磁共振联合使用，胶质瘤的状态是可分的，采用机器学习和模式分类方法可解决胶质瘤分化程度的评价问题，明显优于传统的单模态统计分析方法。

中文关键词： 神经胶质瘤；磁共振；支持向量机；机器学习

英文摘要： Malignant tumor is a serious danger to people life. To evaluate its physiological status and function and understand cancer behaviours is very important to induct the tumor research and clinical remedy. This project researchs on the evaluation of brain glioma function and status based on quantitative MR, and presents a new method to realize clinical diagnosis and grade. Firstly, The project monitors the brain glioma by the perfusion, spectroscopy and DTI MR, and calculates hemodynamics, metabolism and diffusion parameters, which are applied to build the original features to describe glioma status and function. Secondly, the features are optimized based on the support vector machine and data mining, and they have good specificity and sensitivity to brain gliomas.The project evaluates the clinical values of optimized features,and designs new classifiers to bulid the new analysis method for evaluating the glioma function and grade. The project firstly presents an automated method for brain glioma evaluation,which is based on machine learning and pattern classification. There are three contributions. Firstly, a new, automated and standard method for ROI selection is presented. The method reduces the researcher dependency, and provides stable and reliable data for classifier. Secondly, the feature selection is finished by pattern classification,and the classifier for glioma grade is also established, which has good accuracy under test. Finally, a software tool is developed, which includes MR data processing, feature selection and SVM classifier. The tool supports successive research and clinical application. The project proves the glioma status and grade is separable by multi-modal MR. The method based on machine learning and pattern classification can solve the evaluation problem of glioma grade, and it is obviously better than traditional single-modal statistical analysis.

英文关键词： glioma;MR;SVM;machine learning