脑部MRI图像中深层核团的分割方法研究

发布时间：2018-01-15 06:07

本文关键词：脑部MRI图像中深层核团的分割方法研究　出处：《北京工业大学》2015年硕士论文　论文类型：学位论文

【摘要】：临床上对帕金森病、癫痫、阿尔茨海默病等神经系统疾病的治疗主要为使用药物和实施立体定向手术。脑内很多深层核团,如尾状核、壳核、苍白球、丘脑等都是手术中常见的目标靶区。靶区的精确定位是手术成败的关键,决定着手术的最终疗效。因此,提高靶区的定位精度是目前亟待解决的临床问题。本文对磁共振图像(Magnetic Resonance Imaging,MRI)图像中脑内深层核团的自动识别技术进行研究。全面通过挖掘MRI图像脑部结构特点,采用两种研究思路,实现脑内深层核团的精确分割。本文主要研究内容包括:1.实现一种基于配准框架的脑内深层核团的自动分割算法。首先,利用模板图像,将待分割的若干目标子结构视为一个由若干子结构组成的树型结构模板,根据目标识别的难度对模板中各子结构划分等级;其次,依据整合结构模板,建立能够评价目标核团识别效果的能量函数,其各构成项从不同角度描述了衡量识别效果的准则,且融合了各子结构间的相互关联信息;再次,利用基于马尔科夫场(Markov Random Field,MRF)的脑组织分割方法对整合结构模板中的根结构进行分割,并将其作为后续识别过程的初始化信息;最后,利用基于B样条的FFD配准对初步识别的各目标结构进行优化修正。利用本算法对15例MRI图像进行实验,结果表明,尾状核、壳核、苍白球和丘脑结构分割结果与手工分割结果相似度均大于0.75。与其它基于配准框架的分割算法相比,本算法具有更高准确率,分割结果的平均相似度为0.82。本文利用一种获取先验形状模型的新方法,实现在同一框架下依次对多个脑内深层核团的自动分割。其准确率高,无须人工干预,具有较高临床应用价值。2.实现一种基于改进模糊连接度的丘脑及其子结构分割算法。基于在传统模糊连接度框架内增加梯度特征、采用自适应权重的前期工作基础,利用黑白top-hat变换增强图像对比度;结合置信连接度理论,在计算模糊亲和度之前,对目标核团所在感兴趣区域进行自动划分,并计算该区域内灰度与梯度的统计特征。本算法仅需一个种子像素即可自动获取目标感兴趣区,在模糊连接度框架内引入了梯度特征,并可实现权重的自适应调整,减少了人工干预,提高了分割准确性。采用本算法对25例MRI图像进行实验,结果表明,分割结果与手工分割结果相似度均大于0.75。与其它分割算法比较,本文算法在准确率上具有明显优势,同时具有时间代价小、鲁棒性强、主观影响小的优点。本研究实现了基于不同研究思路的脑部MRI图像深层核团的分割方法,可为医生提供更加科学直观的影像学定位参考,为立体定向手术中深层脑部子结构的自动识别提供技术支持。
[Abstract]:Clinical treatment of Parkinson's disease, epilepsy, Alzheimer's disease and other neurological diseases is mainly drug use and stereotactic surgery. Many deep nuclei in the brain, such as caudate nucleus, putamen nucleus, globus pallidus. Thalamus is a common target area in surgery. The accurate location of target area is the key to the success or failure of the operation and determines the final outcome of the operation. It is an urgent clinical problem to improve the accuracy of target location. Magnetic Resonance Imaging is studied in this paper. The automatic recognition technology of deep nuclei in the brain of MRI images was studied. Two kinds of research ideas were adopted by mining the brain structure characteristics of MRI images. The main contents of this paper include: 1. An automatic segmentation algorithm based on registration framework is implemented. First, template image is used. The target substructures to be segmented are regarded as a tree structure template composed of several substructures, and each substructure in the template is classified according to the difficulty of target recognition. Secondly, according to the integrated structure template, the energy function can be established to evaluate the effectiveness of the target nucleus recognition, and its components describe the criteria to measure the recognition effect from different angles. The interrelation information of each substructure is fused. Thirdly, the method of brain tissue segmentation based on Markov Random Random is used to segment the root structure in the integrated structure template. It is used as the initialization information of the subsequent identification process. Finally, the FFD registration based on B-spline is used to optimize the target structure of the initial recognition. 15 cases of MRI images are tested using this algorithm. The results show that the caudate core and shell core are obtained. The similarity between the segmentation results of pallidus and thalamus is greater than that of manual segmentation. Compared with other algorithms based on registration frame, this algorithm has a higher accuracy. The average similarity of segmentation results is 0.82.This paper uses a new method to obtain a priori shape model to realize automatic segmentation of multiple deep nuclei in the same frame. Without manual intervention, it has higher clinical application value. 2. To implement a segmentation algorithm of thalamus and its substructure based on improved fuzzy connectivity, based on adding gradient features in the framework of traditional fuzzy connectivity. The image contrast is enhanced by black and white top-hat transform based on the previous work of adaptive weight. Combining with the theory of confidence connectivity, the region of interest of the target nuclei is automatically divided before the fuzzy affinity degree is calculated. The statistical features of grayscale and gradient in this region are calculated. In this algorithm, only one seed pixel is needed to automatically obtain the region of interest of the target, and the gradient feature is introduced in the framework of fuzzy connectivity. The adaptive adjustment of weight can be realized, the artificial intervention is reduced, and the segmentation accuracy is improved. 25 cases of MRI images are tested by this algorithm, and the results show that. The similarity between the segmentation results and manual segmentation results is greater than 0.75. Compared with other segmentation algorithms, this algorithm has obvious advantages in accuracy, at the same time, it has low time cost and strong robustness. This study has realized the segmentation method of deep nuclei in brain MRI image based on different research ideas, which can provide a more scientific and intuitive imaging location reference for doctors. It provides technical support for automatic recognition of substructure of deep brain in stereotactic surgery.
【学位授予单位】：北京工业大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R445.2;TP391.41

【参考文献】