基于多图谱的多发性硬化病变分割方法研究与应用

发布时间：2018-07-20 09:51

【摘要】：多发性硬化病变是一种比较常见的中枢神经系统疾病,早期症状表现为四肢麻木乏力,后期会导致中风、认知障碍和视力退化等,严重威胁人类的健康。临床上通过核磁共振图像手动勾画病变组织确诊病情,费时费力且具有主观不确定性。因此,研究多发性硬化病变的自动分割算法,提高其分割准确性和稳定性,对于诊断、治疗该疾病具有重要意义。在阅读大量文献的基础之上,本文使用了一种基于多图谱的分割方法,应用于多发性硬化病变的自动分割。研究工作主要有:(1)绪论。介绍了多发性硬化病变分割的研究意义,通过阅读文献,分析了现阶段国内外关于多发性硬化病变分割的研究现状,对本文各部分研究内容作了简要说明。(2)核磁共振图像的分割方法。介绍了脑部病变分割的必备理论知识,主要包括脑部医学图像相关知识和核磁共振成像技术以及核磁共振图像分割方法。对阈值法、聚类法、分水岭算法和小波变换进行了简单介绍。基于图谱的分割方法有三种:基于单图谱的分割方法、基于平均图谱的分割方法和基于多图谱的分割方法。(3)基于多图谱的多发性硬化病变分割算法。对基于多图谱的分割方法流程进行了介绍,并对图像预处理、图像配准与分割以及图像融合这三个步骤进行了详细的介绍。在预处理过程中首先通过BET算法将原始图像去脑壳,然后利用曲线拟合和遗传算法进行偏移场校正。在配准方面,本文主要针对过去配准方法存在精度不高等问题,采用了fsl-anat配准方法。Fsl-anat配准方法利用局部相似性作为配准的相似性测度,并且对变形场施加了有效的约束及平滑可以达到比较好的配准效果。我们以fsl-anat配准的局部相似性作为融合时的权重,它适合于单模态和多模态配准任务。在融合过程中使用了加权选择融合策略。(4)基于多图谱的多发性硬化病变分割方法实验结果分析。利用本文的算法进行实验并实现多发性硬化病变分割。根据脑部病变组织的特征,本文选取了相似性测度来对分割结果进行评价。融合所得到的脑部病变组织在形状和大小方面都和专家手动分割的结果比较接近,十组实验的相似性测度值都达到了0.80以上。说明基于多图谱的多发性硬化病变分割方法,可以高精度地分割出多发性硬化病变组织。
[Abstract]:Multiple sclerosis (MS) is a common disease of central nervous system (CNS). Its early symptoms are numbness and weakness of limbs, which can lead to stroke, cognitive impairment and visual degeneration, which seriously threaten human health. It is time-consuming and laborious and subjective uncertainty to diagnose the disease by manually delineating the pathological tissue by MRI image in clinic. Therefore, it is important for diagnosis and treatment of multiple sclerosis disease to study automatic segmentation algorithm to improve its segmentation accuracy and stability. On the basis of reading a large number of literatures, this paper uses a multi-atlas based segmentation method, which is applied to the automatic segmentation of multiple sclerosis lesions. The main research work is as follows: (1) introduction. This paper introduces the significance of the segmentation of multiple sclerosis, and analyzes the present situation of the segmentation of multiple sclerosis at home and abroad by reading the literature. The research contents of this paper are briefly described. (2) Segmentation method of nuclear magnetic resonance image. This paper introduces the necessary theoretical knowledge of brain lesion segmentation, including related knowledge of brain medical image, magnetic resonance imaging technology and nuclear magnetic resonance image segmentation method. The threshold method, clustering method, watershed algorithm and wavelet transform are briefly introduced. There are three segmentation methods based on map: one based on single map, one based on average spectrum and one based on multi-atlas. (3) multiple sclerosis segmentation algorithm based on multi-atlas. The segmentation process based on multi-atlas is introduced, and the three steps of image preprocessing, image registration and segmentation, and image fusion are introduced in detail. In the process of preprocessing, the original image is removed from the skull by BET algorithm, and then the offset field is corrected by curve fitting and genetic algorithm. In the aspect of registration, aiming at the problem of low precision in the past registration methods, this paper uses the fsl-anat registration method. Fsl-anat registration method uses local similarity as the similarity measure of registration. And the deformation field is subject to effective constraints and smoothing can achieve a better registration effect. We use the local similarity of fsl-anat registration as the weight of fusion, which is suitable for single and multimodal registration tasks. The weighted selection fusion strategy is used in the fusion process. (4) the experimental results of multiple sclerosis segmentation method based on multi-atlas are analyzed. The algorithm of this paper is used to carry out experiments and realize the segmentation of multiple sclerosis disease. According to the characteristics of brain lesions, the similarity measure is selected to evaluate the segmentation results. The shape and size of the brain lesions obtained by the fusion are close to the results of manual segmentation by experts. The similarity measures of the ten groups of experiments are above 0.80. The results show that multiple sclerosis segmentation method based on multi-atlas can be used to segment multiple sclerosis lesions with high accuracy.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R744.51;TP391.41

【参考文献】