基于LGE-MRI的心室瘢痕组织分割系统设计与实现

发布时间：2019-01-05 11:16

【摘要】：目前,全球死亡的主要原因是缺血性心脏病,临床常规的心脏磁共振成像诊断技术,可以全面提供心脏组织关于形态学,组织表征,血流或灌注等信息,LGE-MRI(晚钆增强磁共振成像)是目前用于评估心肌活性的金标准。损伤组织根据T1加权成像,使造影剂出现不同程度的累积导致不同的增强效果,坏死组织具有高信号强度,而心肌的边界几乎不增强。已有研究表明LGE-MRI成像用于指导心肌梗死疾的预后管理治疗的可行性,要实施这些发展起来的成像技术,必须实现心肌梗死区域的重复和可靠分割。因此本文基于具有评估策略的数据集进行心室瘢痕组织分割系统的设计。由于LGE-MRI成像过程会受到潜在的外界噪声干扰,因此图像分割前的平滑预处理步骤十分关键,本文首先对基准图像数据中手动分割出的心肌图像文件和心肌瘢痕图像文件的灰度分布进行特征提取,并对心肌图像应用了高斯滤波、双边滤波和基于形态学开闭运算的噪声平滑技术,得出经形态学滤波处理的图像更能有效保留图像边缘。考虑到获得的数据集中手工划分出的心室内外膜边界区域可能会因为医生先验知识的不足而导致心肌边缘的错分,因此本文对基于灰度修正的图像预处理过程进行了实现,并将灰度修正预处理方法应用于后文提出的区域生长分割,对经灰度修正预处理前后区域生长分割的心肌瘢痕分割结果进行了定量分析,实验验证表明灰度修正能够有效提高分割算法的性能。本文用于系统集成的分割方法主要包括Level Set(水平集)、Graph Cuts(图割)、WaterShed(分水岭)和区域生长分割算法来实现心肌瘢痕组织的分割,每一种分割算法的具体实现均融合了不同的图像预处理过程。针对分割算法性能的评估,本文采用了实验分析的方法,分别从Dice指数、Jaccard系数、Ravd系数和精召率等多个评价指标来对各分割算法的分割结果与手动分割出的金标准进行比较。各分割算法性能评估结果表明,融合了灰度修正的GraphCuts方法与基准分割能够达到更好的一致性。本文分割系统的设计主要包括LGE-MRI图像的文件管理、LGE-MRI图像的降噪预处理、心室瘢痕组织的分割、基于体绘制方法的三维成像显示,以及各分割算法分割结果的定量分析和直方图统计分析。
[Abstract]:At present, the main cause of death in the world is ischemic heart disease, clinical conventional cardiac magnetic resonance imaging diagnosis technology, can provide a comprehensive heart tissue about morphology, tissue characterization, blood flow or perfusion and other information. LGE-MRI (late gadolinium enhanced magnetic resonance imaging) is the current gold standard for assessing myocardial activity. According to T1-weighted imaging of injured tissue, the accumulation of contrast agent in different degree leads to different enhancement effect. The necrotic tissue has high signal intensity, but the boundary of myocardium is hardly enhanced. It has been shown that LGE-MRI imaging is feasible to guide the prognosis management of myocardial infarction disease. In order to implement these developed imaging techniques, repeated and reliable segmentation of myocardial infarction area must be realized. Therefore, the system of ventricular scar tissue segmentation is designed based on the data set with evaluation strategy. Because the LGE-MRI imaging process will be disturbed by the potential external noise, the smoothing preprocessing step before image segmentation is very important. In this paper, firstly, the grayscale distribution of the myocardial image file and the myocardial scar image file are extracted, and Gao Si filter is applied to the myocardial image. Bilateral filtering and noise smoothing technology based on morphological open and close operation show that the image processed by morphological filtering is more effective to preserve the edge of the image. Considering that the boundary region of ventricular inner and outer membrane can be divided manually by the acquired data set, the defect of medical prior knowledge may lead to the misclassification of myocardial edge, so the image preprocessing process based on gray correction is implemented in this paper. The gray correction pretreatment method is applied to the region growth segmentation proposed later. The segmentation results of myocardial scar growth before and after the gray correction pretreatment are quantitatively analyzed. Experimental results show that gray level correction can effectively improve the performance of the segmentation algorithm. In this paper, the segmentation methods for system integration mainly include Level Set (horizontal set), Graph Cuts (map cut), WaterShed (watershed) and region growth segmentation algorithm to achieve myocardial scar tissue segmentation. The implementation of each segmentation algorithm integrates different image preprocessing processes. In order to evaluate the performance of the segmentation algorithm, this paper uses the method of experimental analysis, from the Dice index, Jaccard coefficient, Ravd coefficient and precision rate and other evaluation indicators to compare the segmentation results of each segmentation algorithm with the gold standard of manual segmentation. The performance evaluation results of each segmentation algorithm show that the fusion of gray level modified GraphCuts method and benchmark segmentation can achieve better consistency. The design of this segmentation system mainly includes the file management of LGE-MRI image, the pre-processing of LGE-MRI image, the segmentation of ventricular scar tissue, and the 3D imaging display based on volume rendering method. Quantitative analysis and histogram statistical analysis of segmentation results of each segmentation algorithm.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R445.2;TP391.41

【相似文献】