基于神经网络和图像显著性的心脏图像自动分割算法研究

发布时间：2018-05-28 20:18

本文选题：CT图像 + 心脏分割　；参考：《北京交通大学》2017年硕士论文

【摘要】：心血管疾病目前是导致人类非意外死亡的主要原因之一,近年来呈现逐渐年轻化的趋势,时刻威胁着人类健康生活。心脏是心血管系统的核心器官,获取心脏的生理病理信息是诊断和治疗心血管疾病的关键。CT成像技术具有成像速度快、成像清晰等优点,是心脏检查的常用手段,CT图像的心脏分割对于心血管疾病的诊疗和定量分析具有重要意义。然而,心脏是实体器官,普通的断层图像序列难以完整地呈现出心脏的生理信息,计算机辅助诊断、介入治疗指导、心脏手术导航等技术的实现通常需要得到完整的心脏解剖结构。医学图像的三维可视化是了解心脏解剖结构的主要手段,而图像分割是实现人体器官三维可视化的数据基础。针对心脏图像分割,国内外研究人员提出了多种分割方法,但是大部分算法是针对心房和心室的分割,并不能满足全心脏分割的要求。同时,心脏的搏动和血液流动过程中得到的CT图像容易出现伪影、边界弱化等问题,影响心脏壁的分割效果,传统手动分割方法需要医务人员的大量经验,分割效率低。因此全心脏的自动分割一直是医学图像处理研究领域具有挑战性的热点问题。本文以医学图像的分割和识别为基础,针对切片中心脏图像与其它组织区别明显,以及心脏CT图像序列中相邻切片间有很高相似性的特点,提出了基于神经网络和图像显著性的全自动分割算法,并对分割后的心脏图像进行了三维可视化重建。本文的主要内容及创新点:1.利用视觉显著性技术,实现基于CT图像的心脏分割。采用了一种图像显著性检测算法,对CT图像进行显著性计算,最后利用显著图像对原CT图像进行分割,得到完整的心脏图像。2.将分割任务分解为定位和分割两部分,采用卷积神经网络和堆叠降噪自编码网络分别进行实现。构建卷积神经网络,实现其对图像中心脏的定位功能,利用定位结果对原心脏CT图像进行裁剪,去掉部分非目标区域;构造了一个堆叠降噪自编码网络,利用手动分割图像对网络进行训练,实现网络对心脏CT图像中属于心脏组织的像素点的分类识别,最后基于分类结果实现对心脏图像的分割。3.对上述分割算法分割后的结果,分别与人工分割结果进行定量评价分析,并采用面绘制和体绘制两种方法对分割结果进行可视化重建。
[Abstract]:Cardiovascular disease is one of the main causes of human unaccidental death. In recent years, cardiovascular disease is becoming younger and threatening human health. The heart is the core organ of the cardiovascular system. Obtaining the physiological and pathological information of the heart is the key to diagnose and treat cardiovascular diseases. Ct imaging technology has the advantages of fast imaging speed and clear imaging. The heart segmentation of CT images is of great significance for the diagnosis, treatment and quantitative analysis of cardiovascular diseases. However, the heart is a solid organ, the common image sequence of tomography is difficult to show the physiological information of the heart, computer-aided diagnosis, interventional therapy guidance, The complete anatomical structure of the heart is usually required for the realization of cardiac surgery navigation and other techniques. Three-dimensional visualization of medical images is the main means to understand the anatomical structure of the heart, and image segmentation is the data base to realize three-dimensional visualization of human organs. For heart image segmentation, researchers at home and abroad have proposed a variety of segmentation methods, but most of the algorithms are aimed at the atrial and ventricular segmentation, which can not meet the requirements of the whole heart segmentation. At the same time, the CT images obtained in the process of heart beating and blood flow are prone to artifact and boundary weakening, which affect the segmentation effect of heart wall. Traditional manual segmentation methods need a lot of experience of medical personnel, and the efficiency of segmentation is low. Therefore, automatic heart segmentation has been a challenging hot issue in the field of medical image processing. Based on the segmentation and recognition of medical images, this paper aims at the obvious difference between the heart image and other tissues in the slice and the high similarity between the adjacent slices in the cardiac CT image sequence. An automatic segmentation algorithm based on neural network and image saliency is proposed, and the segmented heart image is reconstructed by 3D visualization. The main content and innovation of this paper: 1. Using visual salience technology, the heart segmentation based on CT image is realized. An image saliency detection algorithm is used to calculate the significance of CT image. Finally, the original CT image is segmented by salient image, and the complete heart image is obtained. The segmentation task is divided into two parts: location and segmentation. Convolution neural network and stacking noise reduction self-coding network are used to realize the segmentation task. A convolutional neural network is constructed to locate the heart in the image. The original heart CT image is clipped using the localization results, and some non-target regions are removed, and a stack noise reduction self-coding network is constructed. The network is trained by manual segmentation image to realize the classification and recognition of the pixels belonging to the heart tissue in the cardiac CT image. Finally, the segmentation of the heart image is realized based on the classification results. 3. The segmentation results of the above segmentation algorithms are evaluated and analyzed quantitatively with the artificial segmentation results, and the segmentation results are reconstructed visually by surface rendering and volume rendering.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP183

【相似文献】