MR数据的贝叶斯空间统计学方法及其对胶质瘤的立体计量分析

项目名称： MR数据的贝叶斯空间统计学方法及其对胶质瘤的立体计量分析

项目编号： No.81202284

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 预防医学、地方病学、职业病学、放射医学

项目作者： 赵倩

作者单位： 广州医科大学

项目金额： 23万元

中文摘要： 基于空间相关性的空间统计学，是现代统计学的研究热点。研究多模MR数据的空间统计学方法，能促进新的多元信息融合理论的发展。然而，由于MR数据自身的特点，不同扫描序列的空间分辨率通常不同；需要与实际相符的先验信息，提高模型的准确度和精度；由于肿瘤处于中枢神经系统，难以采集病理标本验证模型的拟合效果。基于病理数据为先验信息的贝叶斯空间统计方法有望解决这些问题。本项目拟采用贝叶斯空间统计方法实现多元MR属性对胶质瘤空间分布的解释和预测，并使模型具有一定稳健性和自适应性的优点。具体开展以下研究：①以病理结果为先验知识，引入核函数构造贝叶斯模型先验分布；②多模MR数据空间位置配准；③多模MR数据的多尺度模型的量化分析；④肿瘤浸润范围的空间模型构建和验证。项目的研究成果不仅对胶质瘤的定量分析有重要的临床意义，而且还将空间统计学的方法拓展到医学领域，同时给医学影像诊断提供了一种新的研究思路。

中文关键词： 空间统计学；贝叶斯方法；核估计；广义线性混合模型；

英文摘要： Spatial statistics based on spatial correlation is a hot issue of advanced statistics. Research on spatial statistical approach for multi-modality MR data would promote the new information fusion theory. However,due to the characters of MR data, there are different spatial resolutions for different MR sequences. Prior information, which is consistent with the fact, is required for the precision and accuracy of statistical model. Furthermore, the tumor is located within the central nervous system. Therefore, it is difficult to obtain pathology panel and verify the modeling effect. Bayesian statistical method, relying on phathology as prior information, will be a hopeful way to solve these problems. This project would describe the spatial structure of gliomas with Bayesian statistical method, and construct spatial auto-regression model with multi-modality MR data to forecast the spatial distribution of gliomas. Meanwhile, the model would have the virtues of robustness and adaptability. This work is based on establishment of a glioma mode in rat as follows;① Construction of the prior distribution of Bayesian model with phathology and kernel function;②Registration for spatial position of multi-modality MR data; ③ Multi-scale quantitative analysis for multi-modality MR data; ④ Modeling and verification for invasive e

英文关键词： Spatical statistics；Bayses method；kernel estimate；general linear mixed model；