基于概率图谱引导的群组自适应时序脑MR图像脑提取方法研究

项目名称： 基于概率图谱引导的群组自适应时序脑MR图像脑提取方法研究

项目编号： No.U1504606

项目类型： 联合基金项目

立项/批准年度： 2016

项目学科： 无线电电子学、电信技术

项目作者： 王雅萍

作者单位： 郑州大学

项目金额： 27万元

中文摘要： 脑提取是神经影像学研究中脑图像处理的第一步，随着多样化大型数据集的涌现，广泛适用于多种数据集的精确且鲁棒脑提取方法亟待提出。为了更进一步地发现时序脑图像中所发生的动态变化，需要开发一种精确一致的脑提取方法用于敏感捕捉这种细微变化。本项目拟提出一种脑概率图谱作为先验知识引导的群组自适应时序脑提取方法。方法分两个阶段进行，通过学习先验知识构建脑概率图谱，并与基于变形模型的精确修正相结合以提高算法鲁棒性与精确度；针对时序图像脑提取时间一致性的问题，通过群组配准使各时间点图像对应起来并建立相同的初始表面，通过加入时间平滑约束，驱动各时间点上表面模型同步进行变形，获取精确一致的时序脑提取结果。拟研究的基于先验知识引导的群组自适应时序脑提取方法将提高时序脑MR图像脑提取的精确性和一致性，为辅助临床脑疾病诊断和探索人脑的基本工作原理等奠定良好的基础。申请人有多年脑图像处理和分析的工作基础。

中文关键词： 医学图像处理；核磁共振成像；脑提取；时序脑图像；脑概率图谱

英文摘要： Brain extraction is the first step of brain image processing in neuroscience research. Robust brain extraction image which is applicable to diverse large-scale dataset is needed. To reveal subtle but dynamic changes in longitudinal brain images, robust and consistent brain extraction method is urgently needed. The proposal aims to develop a probability map guided group adaptive brain extraction approach of longitudinal brain MR images. The project will be executed in two stages. The brain probability map is first constructed and then employed in a deformable model for refining the extraction result. For longitudinal brain images, groupwise registration is performed and the same initial surface meshes are placed on images of all time points. With the temporal smoothness constraint, all the surface meshes are simultaneously driven to evolve to the respective target brain boundaries by the surface model. Accurate and longitudinal consistent results will be achieved. The robustness and consistency of the proposed brain extraction method will provide an important foundation for patient diagnosis and for exploring how our brain works. The applicant has worked in the field of brain image analysis for more than 5 years.

英文关键词： Medical Image Processing;MRI;Brain extraction;Longitudinal Brain Images;Brain Probability Map