基于模糊数学的肺部CT图像处理方法的研究
VIP免费
摘 要
肺癌是当今对人类健康与生命危害最大的恶性肿瘤之一，对肺癌的早期发现
与治疗是提高肺癌病人存活率的关键。基于 CT 图像的肺结节计算机辅助检测
（Computer Aided Detection, CAD）技术可以大大提高肺癌的诊断准确率与工作效
率，是目前生物医学工程领域研究的热点之一。
CT 图像是肺癌诊断和治疗的重要依据，它使医生对肺内病变部位的观察更直
接、更清晰，确诊率也更高，在临床上具有非常重要的应用价值。图像天生具有
模糊性，肺 CT 图像在获取过程中，由于肺部解剖结构的复杂性、形状的不规则性、
部分容积效应以及噪声的影响等，造成内在的不确定性或模糊性。模糊数学是专
门研究和分析具有模糊性质的自然现象的学科，特别适合于处理不确定性问题。
因此，将模糊数学方法应用于肺部 CT 图像处理有其内在的合理性和必然性。
本文在前人研究工作的基础之上，做了以下工作：
(1) 本文对近年来基于模糊数学的医学图像处理与分析方法进行了综述。介绍
了模糊集合、模糊聚类、模糊推理、模糊连接度、模糊计算智能等方法在医学图
像处理和分析中的应用，并对不同方法的原理、优缺点进行了讨论。研究表明，
模糊数学在医学图像处理方面已经取得了大量的应用成果。但作为一门年轻学科，
模糊数学在医学图像处理与分析中的应用仍然有很大的发展空间。随着计算机技
术的不断成熟和新技术的不断涌现，基于模糊数学的医学图像处理与分析方法也
将继续发展和完善。
(2) 本文对模糊增强非实质性肺结节进行了实验研究。非实质性肺结节因密度
较低及边缘不清晰等原因成为肺结节检测的一个难点。在参考大量相关文献的基
础上，尝试了经典的 Pal-King 模糊图像增强算法及其自适应选取、改进的单层次
和基于模糊对比度的三种改进的方法，之后给出了适用于非实质肺结节增强的改
进算法。实验结果证明，提出的方法能够在降低血管干扰的同时，有效地增强非
实质性肺结节的边界，提高与肺实质的对比度。本文研究的模糊增强非实质性肺
结节工作可为后续对结节的分割及检测奠定基础。
(3) 本文对基于模糊聚类的亚实质性肺结节分割进行了研究。肺结节的精确分
割是基于 CT 图像的肺结节计算机辅助检测的关键步骤。针对肺 CT 图像所具有的
模糊性质及亚实质性肺结节的特点，通过对模糊 C均值聚类（FCM）及其改进算
法核模糊聚类（KFCM）和加权核模糊聚类（WKFCM）进行实践，提出一种改进
的利用血管及类别结构信息加权的适用于亚实质肺结节的核模糊聚类（IWKFCM）
三维分割方法。本文分别采用 18 个临床数据和 36 个LIDC 标准数据对所提出的分
割方法进行实验，以放射科医生手动分割的区域作为金标准计算重合率，平均值
分别为 71.65%及76.18%，且假阳性率及假阴性率均低于 17%。实验结果表明，相
较于 FCM，KFCM 与WKFCM，IWKFCM 能够获得更准确的分割结果，并且分割
效果同时优于目前文献报道的其他非模糊数学方法。本文方法为 CAD 系统提供了
一种分割亚实质性肺结节的工具。
本论文所研究内容可为基于 CT 图像的肺结节计算机辅助检测系统后续对肺
结节的特征提取及分类奠定基础。
关键词：计算机辅助检测 肺结节 模糊数学 模糊增强 改进的加
权核模糊聚类 三维分割
ABSTRACT
Lung cancer is the leading cause of cancer death among people nowadays.
Computer-Aided Detection (CAD) for pulmonary nodules on CT images can help
radiologist find pulmonary nodules efficiently and decrease the rate of missed diagnosis,
which provides a significant way for lung cancer diagnosis in early stage.
CT image is an important basis in lung cancer diagnosis and treatment. It makes
the doctor to observe the lung tissues more direct and clear, then to improve accuracy
rate in diagnosis. So CT images have very important value in clinical application.
Generally, images suffer information loss when the three dimensions target is projected
on the two dimensions image plane. There is some confusion in defining boundary,
region and texture of image, bringing fuzzy character when explaining the result of the
low level image processing. All in all images are fuzzy by nature, so does the lung CT
image. Fuzzy mathematics is a branch of mathematics subjects which studies the fuzzy
phenomena and their concepts, now is used in wide areas. Therefore the applications of
fuzzy mathematics methods in medical image processing and analysis are rational and
necessary.
Based on consulting much previous research studies, this paper has finished some
work as following:
(1) A detailed review about the fuzzy mathematics methods in medical image
processing and analysis during the past few years is done in this paper. The applications
of fuzzy sets, fuzzy clustering, fuzzy inference, fuzzy connectedness and fuzzy
computing intelligence method in medical image processing and analysis are introduced,
with discussions of these different methods on their principles, merit and demerit,
characteristics and existing problems. Researches show that fuzzy mathematics has
already made a large amount of fruitful applications results in the area of medical image
processing. As a young subject, it still has shown a huge potential in the application of
medical image processing and analysis. With the further development of computer
technology and the continuous emergence of advanced technology, medical image
processing and analysis based on fuzzy mathematics will become mature and perfect.
(2) An experimental investigation of fuzzy enhancement algorithms for non-solid
pulmonary nodules is performed in this paper. For the detection of the non-solid
pulmonary nodules with irregular shape, blurry edge and low contrast to lung
parenchyma is more difficult, the image enhancement for non-solid pulmonary is
necessary in CAD systems. Based on consulting a lot of related references, we attempt
the classic Pal-King fuzzy enhancement algorithm and several kinds of improved
algorithms. Finally an applicative method for the enhancement of non-solid pulmonary
nodules is proposed. Practices prove that our proposed method can efficiently enhance
the non-solid pulmonary nodules’ contrast to lung parenchyma and decrease the
disturbance of blood vessels at the same time.
(3) Segmentation is a key step in CAD systems, for its accuracy directly affects the
following steps of CAD system. According to the characteristics of sub-solid pulmonary
nodules in lung CT images, through attempting fuzzy c-means (FCM), its improved
algorithms kernel fuzzy c-means (KFCM) and weight kernel fuzzy c-means (WKFCM),
a novel weighed kernel fuzzy c-means (IWKFCM) method considering information
about vessels structure and classes’ distribution as weights is proposed in this paper. The
method is tested on both clinical data and standard LIDC datasets including 18 clinical
nodules and 36 LIDC nodules. Overlaps of two datasets are 71.65% and 76.18%
respectively, both with low FPR and FNR less than 17%. Experimental results show that
the proposed method can get more accurate result than FCM, KFCM and WKFCM, and
also is superior to other methods published in recent reports and literature.
Our research work in this paper can provide a consultative reference to the further
step of feature extraction and classification for pulmonary nodules in CAD systems.
Key Words：Computer-Aided Detection， Pulmonary nodule, Fuzzy
mathematics， Fuzzy enhancement， Improved weighed kernel fuzzy
c-means，3D segmentation
目 录
中文摘要
ABSTRACT
第一章 绪 论 .................................................................................................................. 1
§1.1 问题描述 ........................................................................................................... 1
§1.2 本文主要研究内容 ........................................................................................... 3
§1.3 本文结构 ........................................................................................................... 4
第二章 模糊数学医学图像处理与分析方法 ................................................................ 5
§2.1 模糊数学理论基础 ........................................................................................... 5
§2.2 医学图像处理与分析概述 ............................................................................... 6
§2.3 基于模糊数学的医学图像处理与分析 ........................................................... 8
§2.3.1 Pal-King 算法 .......................................................................................... 9
§2.3.2 模糊聚类 ............................................................................................... 10
§2.3.3 模糊推理 ................................................................................................11
§2.3.4 模糊连接度 ........................................................................................... 12
§2.3.5 模糊计算智能 ....................................................................................... 14
§2.3.6 其他模糊数学方法 ............................................................................... 18
§2.4 小结 ................................................................................................................. 19
第三章 基于模糊数学的非实质肺结节增强研究 ...................................................... 20
§3.1 引言 ................................................................................................................. 20
§3.2 传统图像增强方法 ......................................................................................... 22
§3.3 经典的 Pal-King 模糊图像增强算法 ............................................................. 23
§3.4 Pal-King 算法的发展与改进 .......................................................................... 25
§3.4.1 基于自适应阈值选取的模糊增强算法 ............................................... 25
§3.4.2 改进的单层次模糊增强算法 ............................................................... 26
§3.4.3 基于模糊对比度的图像增强方法 ....................................................... 27
§3.5 改进的适用于非实质肺结节的模糊增强算法 ............................................. 28
§3.6 实验结果与分析 ............................................................................................. 30
§3.7 小结 ................................................................................................................. 31
第四章 基于改进的加权核模糊聚类亚实质肺结节三维分割研究 .......................... 33
§4.1 引言 ................................................................................................................. 33
§4.2 传统图像分割方法 ......................................................................................... 35
§4.3 模糊聚类算法 ................................................................................................. 36
§4.3.1 经典模糊 C均值(FCM)聚类................................................................ 37
§4.3.2 模糊 C均值聚类的改进....................................................................... 38
§4.4 改进的基于血管及类别结构加权的核模糊聚类算法 ................................. 39
§4.4.1 核模糊 C均值(KFCM)聚类................................................................. 39
§4.4.2 基于血管及类别结构加权的核模糊聚类(IWKFCM)算法 ................ 41
§4.5 基于改进的加权核模糊聚类亚实质肺结节三维分割 ................................. 43
§4.5.1 ROI 选取 ............................................................................................... 43
§4.5.2 加权核模糊聚类 ................................................................................... 43
§4.5.3 三维连通域标记及形态学后处理 ....................................................... 44
§4.6 实验结果与分析 ............................................................................................. 44
§4.6.1 临床数据实验 ....................................................................................... 45
§4.6.2 LIDC 标准数据集实验 ......................................................................... 50
§4.6.3 实验结果对比 ....................................................................................... 55
§4.7 小结 ................................................................................................................. 55
第五章 总结与展望 ...................................................................................................... 57
§5.1 总结 ................................................................................................................. 57
§5.2 展望 ................................................................................................................. 57
参考文献 ........................................................................................................................ 59
在读期间公开发表的论文和承担科研项目及取得成果 ............................................ 65
致 谢 .............................................................................................................................. 66
相关推荐
作者：牛悦
分类：高等教育资料
价格：15积分
属性：70 页
大小：3.14MB
格式：PDF
时间：2025-01-09
