MRI脑部组织分割方法研究
发布时间：2018-08-07 14:01
【摘要】：核磁共振成像具有软组织成像效果好、空间分辨率高、非介入性、扫描角度灵活等优点,已成为脑疾病诊断的重要辅助手段。准确分割脑部组织对后续解剖脑部疾病,诸如阿尔茨海默病、多发性硬化症、帕金森以及精神分裂症的分析与研究具有重要的指导意义。由于脑组织物理属性不同,在MR图像上呈现不同的灰度范围,高斯混合模型(Gaussian Mixture Model,GMM)已成为一种描述灰度缓慢变化的理想模型。但是传统的GMM以像素的独立性假设为前提,组织的空间结构信息往往被忽略。同时,由于脑组织自身解剖结构的复杂性,加上成像过程当中出现的偏移场、部分容积效应、噪声等物理性因素,导致MR图像的分段常量性被破坏。为了提高脑组织分割精度,本文重点研究了基于组织概率图谱先验信息和后验邻域信息的高斯混合模型脑组织三维分割算法。具体研究内容如下:1、提出了一种基于图谱先验信息的高斯混合模型脑组织三维分割算法(PA-GMM)。实验结果表明PA-GMM算法可以解决传统GMM由于空间信息缺失而导致在噪声和偏移场增大情况下误分率提高的情况,有效提高了脑组织分割精度。2、MR图像当中存在偏移场,而当偏移场过大时候,会严重影响最后算法的分割精度。因此,本文在PA-GMM基础之上,实现了一种基于偏移场校正的PA-GMM算法。实验结果表明,该方法可以快速有效对MR图像进行3D分割。通过分割与偏移场校正交替迭代进行,比传统预处理阶段先进行偏移场校正,然后再进行组织分割的效果要好。3、为了进一步提高算法在高噪声下的分割精度,利用后验概率的邻域信息和脑组织概率图谱的空间解剖结构先验信息,重新设计混合系数的表达方式,提出了一种SNPA-MGMM分割算法。该算法不仅能够在抑制噪声方面上表现突出,而且能够分割像GM和CSF这样的复杂重叠区域,并且能够保留边缘细节信息。4、本文主要采用BrainWeb的模拟数据集和IBSR的两组真实数据集(v1.0和v2.0)作为测试数据,并将改进后的算法与一些最新文献和医学软件上的分割结果进行对比,最后利用专家手动分割的结果(俗称金标准)进行定量分析与比较。实验结果表明,本文提出的方法可以有效提高组织分割精度。
[Abstract]:Magnetic resonance imaging (MRI), which has the advantages of good soft tissue imaging, high spatial resolution, non-interventional and flexible scanning angle, has become an important auxiliary method for the diagnosis of brain diseases. Accurate segmentation of brain tissue is of great significance in the analysis and research of subsequent anatomical brain diseases such as Alzheimer's disease, multiple sclerosis, Parkinson's disease and schizophrenia. Because of the different physical properties of brain tissue, the Gao Si hybrid model (Gaussian Mixture model has become an ideal model for describing the slow change of gray scale. However, the traditional GMM is based on the assumption of pixel independence, and the spatial structure information of the organization is often ignored. At the same time, due to the complexity of the anatomical structure of brain tissue, the offset field, partial volume effect, noise and other physical factors in the imaging process, the segmental constant of Mr image is destroyed. In order to improve the accuracy of brain tissue segmentation, this paper focuses on the 3D segmentation algorithm of Gao Si mixed model based on prior information of tissue probability map and posteriori neighborhood information. The main contents are as follows: 1. A Gao Si hybrid model of brain tissue segmentation algorithm (PA-GMM) based on the prior information of the map is proposed. The experimental results show that the PA-GMM algorithm can solve the problem that the misdivision rate increases in the case of increased noise and offset field caused by the absence of spatial information in the traditional GMM, and can effectively improve the segmentation accuracy of brain tissue. 2. There exists an offset field in the brain tissue segmentation accuracy. When the offset field is too large, it will seriously affect the segmentation accuracy of the final algorithm. Therefore, a PA-GMM algorithm based on offset field correction is implemented on the basis of PA-GMM. Experimental results show that the proposed method can be used to segment Mr images quickly and effectively. By alternating iteration of segmentation and offset field correction, the effect of migration field correction is better than that of traditional preprocessing stage, and then the effect of tissue segmentation is better. In order to further improve the segmentation accuracy of the algorithm under high noise, Using the neighborhood information of posterior probability and the prior information of spatial anatomical structure of brain tissue probability map, a new SNPA-MGMM segmentation algorithm is proposed by redesigning the expression of mixed coefficients. The algorithm can not only suppress noise, but also segment complex overlapping regions such as GM and CSF. And can keep edge detail information. 4. This paper mainly uses BrainWeb's simulated data set and IBSR's two groups of real data sets (v1.0 and v2.0) as test data, and compares the improved algorithm with some new literature and medical software segmentation results. Finally, the expert manual segmentation results (commonly known as gold standard) for quantitative analysis and comparison. Experimental results show that the proposed method can effectively improve the accuracy of tissue segmentation.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R445.2;TP391.41
[Abstract]:Magnetic resonance imaging (MRI), which has the advantages of good soft tissue imaging, high spatial resolution, non-interventional and flexible scanning angle, has become an important auxiliary method for the diagnosis of brain diseases. Accurate segmentation of brain tissue is of great significance in the analysis and research of subsequent anatomical brain diseases such as Alzheimer's disease, multiple sclerosis, Parkinson's disease and schizophrenia. Because of the different physical properties of brain tissue, the Gao Si hybrid model (Gaussian Mixture model has become an ideal model for describing the slow change of gray scale. However, the traditional GMM is based on the assumption of pixel independence, and the spatial structure information of the organization is often ignored. At the same time, due to the complexity of the anatomical structure of brain tissue, the offset field, partial volume effect, noise and other physical factors in the imaging process, the segmental constant of Mr image is destroyed. In order to improve the accuracy of brain tissue segmentation, this paper focuses on the 3D segmentation algorithm of Gao Si mixed model based on prior information of tissue probability map and posteriori neighborhood information. The main contents are as follows: 1. A Gao Si hybrid model of brain tissue segmentation algorithm (PA-GMM) based on the prior information of the map is proposed. The experimental results show that the PA-GMM algorithm can solve the problem that the misdivision rate increases in the case of increased noise and offset field caused by the absence of spatial information in the traditional GMM, and can effectively improve the segmentation accuracy of brain tissue. 2. There exists an offset field in the brain tissue segmentation accuracy. When the offset field is too large, it will seriously affect the segmentation accuracy of the final algorithm. Therefore, a PA-GMM algorithm based on offset field correction is implemented on the basis of PA-GMM. Experimental results show that the proposed method can be used to segment Mr images quickly and effectively. By alternating iteration of segmentation and offset field correction, the effect of migration field correction is better than that of traditional preprocessing stage, and then the effect of tissue segmentation is better. In order to further improve the segmentation accuracy of the algorithm under high noise, Using the neighborhood information of posterior probability and the prior information of spatial anatomical structure of brain tissue probability map, a new SNPA-MGMM segmentation algorithm is proposed by redesigning the expression of mixed coefficients. The algorithm can not only suppress noise, but also segment complex overlapping regions such as GM and CSF. And can keep edge detail information. 4. This paper mainly uses BrainWeb's simulated data set and IBSR's two groups of real data sets (v1.0 and v2.0) as test data, and compares the improved algorithm with some new literature and medical software segmentation results. Finally, the expert manual segmentation results (commonly known as gold standard) for quantitative analysis and comparison. Experimental results show that the proposed method can effectively improve the accuracy of tissue segmentation.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R445.2;TP391.41
【参考文献】
相关博士学位论文 前4条
1 杨红U，
本文编号：2170251
本文链接：https://www.wllwen.com/yixuelunwen/fangshe/2170251.html
