【作者】 颜洁；

【导师】 贺士娟；

【作者基本信息】 河北工业大学 ， 微电子学与固体电子学， 2003， 硕士

【摘要】 脑电图EEG和脑磁图MEG研究属于无创伤性的脑功能成像技术，主要研究由头皮测得的脑电或脑磁信号如何获取脑内源定量的三维信息。构造与实际情况相匹配的头模型是求解脑电正、逆问题的先决条件和临床有效性的关键。进行头模型构建的前提就是人头部磁共振图像脑轮廓的提取和分割问题。由于头内部组织结构复杂，因此获取人脑体数据的分割算法是比较困难的。在我们的研究工作中，采用了可变形模型方法。近年来，可变形模型鉴于其良好的性能，在医学图像分割中得到了广泛的应用。本文对可变形模型的基本概念和研究进展进行了介绍，给出了其应用于头部磁共振图像提取脑区域的算法实现，并针对我们在实际研究工作中遇到的问题，进行了相应的改进，取得了良好的效果。本文的研究工作主要包括以下几个方面：针对GVF snake算法提取脑轮廓时不能很好的收敛于光滑性差的脑轮廓的问题，对GVF snake算法进行了改进，构建了一动态变化矢量场，把GVF场的约束权值设为随目标特征自适应变化，从而提出了一种新的算法，称之为AGVF(adaptable gradient vector flow)snake。对于可变形模型对初始轮廓的依赖较大，外力的作用范围有限，迭代算法收敛时间较长等问题，提出了一种新的外力，我们称之为基于区域的可变膨胀力(Region based variable expansion force)。在可变形模型的成长过程，外力在不同阶段与区域特征进行有针对性的结合，从而较好的解决了原有算法的缺陷。在提取人头部磁共振图像T2加权像的横断面的脑轮廓时，因为在图像内部存在许多的局部极小值，严重的影响了算法向全局极小收敛。为了解决这个问题，我们在可变形模型中加入随机扰动因素，考虑在迭代过程引入模拟退火的思想，提出了一种称之为基于模拟退火的可变形模型SA snake(simulated annealing snake)的新算法，算法具有全局搜索能力。更多还原

【Abstract】 EEG (electroencephalography) and MEG (magnetoencephalongraphy) are noninvasive methods for exploring the neural electrical activities of brain. Brain segmentation is a very important and difficult part in EEG/MEG research, and many researchers have been working on this field.To constructing real head models , the problem of brain contour finding from head MRI images must be solved. Due to the complex composition of the head ,it is difficult to segment brain from the head. The deformable model algorithm is adopted in this research.Deformable models ,also known as ’snakes’, have been one of the most active and successful research areas in image segmentation. Deformable models are now extensively used in medical image analysis to locate and depict tissues because of its good boundary integration and features extraction. The idea of deformable models are introduced . The research , development and applications of the deformable models are reviewed . The algorithm is realized to extract brain contour from MRI head images . Some improvement on deformable models is presented . The proposed algorithms have good performances.Some improvement on the GVF snake algorithm is presented to solve the problem of GVF snake’s inability to converge to the rough contour of the brain. A variable vector field is designed and the constraint parameters of the GVF field is made adaptable to the target object features. The proposed algorithm is called AGVF snake algorithm.When deformable models are used to extract brain contour from the head MR images, problems associated with initialization and time consuming iteration limit their utility. To overcome these problems, a region based variable expansion force is proposed . During the growing period of the deformable models , external forces vary according to the region information . The new algorithm better the performance of the deformable models .When deformable models are used to extract brain contour from head MRI T2 images ,local minimums largely affect the algorithm’s converging to the global minimum. The stochastic disturbance is brought to the deformable models and the simulated annealing isintroduced to the iteration process. A new algorithm named SA snake (simulated annealing snake) is proposed . The algorithm has the ability to search globally.更多还原