多阈值与多图谱方法在医学图像分割上的应用研究

发布时间：2017-12-26 18:32

本文关键词：多阈值与多图谱方法在医学图像分割上的应用研究　出处：《山东大学》2017年硕士论文　论文类型：学位论文

【摘要】：医学图像是医疗人员进行疾病诊断、风险评估及制定治疗计划的重要辅助性工具,随着成像技术的快速发展,海量医学图像数据的产生给计算机辅助的医学图像处理技术提出了新要求。图像分割能够对图像中的每个像素点进行类别划分,为医疗人员提供准确的解剖学信息,如病灶的位置、大小、形状信息,同时为其他图像处理技术,如三维重建,提供信息输入。因此,围绕医学图像展开的三维重建、可视化、定量分析等工作的质量很大程度上取决于医学图像分割的质量。正是出于这样的原因,医学图像分割具有极大的研究价值。近30年间,医学图像分割领域产生了众多经典算法,常用的基于灰度、基于可变模型的分割方法能够对肺部、心脏、乳腺、肾脏等多种器官组织进行高精度的自动分割,极大地简化了医疗人员的工作,也提高了患者康复的几率。但是,对于一些特定部位、特定模态的图像,目前仍无令人满意的分割方法。例如,在脑MR图像分割过程中,由于核磁共振成像设备的电子噪声所造成的灰度不均匀及设备分辨率限制造成的容积效应,不同脑组织呈现出极其相近的灰度,且各组织边缘重叠交错,给基于灰度和基于可变模型的分割方法带来了极大的困难;在主动脉CT图像成像时,主动脉区域与周围组织器官呈现出相近的灰度、纹理特征,纤薄的血管壁也无法提供清晰的边缘,因而,灰度、边缘、纹理等特征在主动脉分割时无法提供有用信息。基于以上的原因,针对脑MR图像存在的偏场效应等问题,本文使用了信息学中的模糊熵,引入隶属度函数的概念,将图像中的像素按照多个隶属度函数的值"软性"地划分到各类,克服了传统阈值分割方法"硬性"分割难以解决偏场效应的不足。在对模糊熵函数进行优化时,本文对回溯搜索算法进行了针对性的改进,增强了其局部搜索能力,能够更精确、更高效地得到最佳的多个阈值。实验结果表明,本文的方法比传统的Otsu和Kapur阈值法更加有效,且鲁棒性更高,对偏场效应有更好的抑制作用。针对主动脉CT图像中存在的区分性不高的问题,本文使用了基于图谱的分割方法,并且使用了一种联合标签融合策略,在考虑图谱与目标图像关系的同时,也兼顾了各图谱之间的相互关系,克服了图谱冗余信息的干扰,从而提高了对主动脉区域的分割精度。实验结果表明,本文提出的主动脉分割方法能够提供90%左右的分割精度,且分割性能和鲁棒性高于结合了其他标签融合策略的图谱分割法。
[Abstract]:Medical image is an important auxiliary tool for medical staff to diagnose diseases, assess risks and formulate treatment plans. With the rapid development of imaging technology, the emergence of massive medical image data has brought new demands for computer-aided medical image processing technology. Image segmentation can classify every pixel in the image, provide accurate anatomical information for medical staff, such as location, size and shape information of the lesion, and provide information input for other image processing technologies, such as 3D reconstruction. Therefore, the quality of medical image segmentation depends largely on the quality of the 3D reconstruction, visualization and quantitative analysis around medical images. It is for this reason that medical image segmentation has great research value. The past 30 years, the field of medical image segmentation has produced many classic algorithms, automatic segmentation, segmentation method based on the commonly used gray model based on variable precision can be obtained in the lung, heart, kidney, breast and other tissues, which greatly simplifies the medical staff, but also improve the chances of rehabilitation. However, there are still no satisfactory segmentation methods for some specific parts and specific modal images. For example, in the segmentation of brain MR image process, due to the volume effect of gray level noise electronic magnetic resonance imaging equipment caused by non-uniform resolution and equipment limitations caused by different brain tissues, showing a gray very similar, and each edge of overlapping organizations, based on the gray segmentation method based on deformable model and brought great the difficulties in CT imaging; aortic, aortic region and surrounding tissues and organs showed gray and texture features are similar to that of the vascular wall slim cannot provide clear edge, therefore, gray, edge and texture features can provide useful information on the aortic segment. Based on the above reasons, the existence of brain MR image bias field effect problem, this paper uses the fuzzy entropy in Information Science, the membership function is introduced, the pixels in the image according to a plurality of membership function values of "soft" is divided into various types, to overcome the traditional threshold segmentation method "" it is difficult to solve the shortage of hard segmentation bias field effect. When optimizing the fuzzy entropy function, we improve the backtracking algorithm, enhance its local search ability, and get the best multiple thresholds more accurately and efficiently. The experimental results show that the proposed method is more effective than the traditional Otsu and Kapur threshold method, and is more robust, and has a better inhibition effect on partial field effect. In view of the existing image in aortic CT distinction is not high, this paper use the segmentation method based on mapping, and the use of a joint tag fusion strategy, considering the relationship between map and target image at the same time, taking into account the relationship between the atlas, to overcome the interference of redundant information, so as to improve the the aortic area segmentation accuracy. The experimental results show that the proposed aortic segmentation method can provide about 90% of the segmentation accuracy, and the segmentation performance and robustness are higher than those combined with other tag fusion strategies.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R319;TP391.41

【相似文献】