当前位置:主页 > 医学论文 > 妇产科论文 >

宫颈细胞学涂片自动判读方法研究

发布时间:2018-04-26 02:23

  本文选题:医学图像 + 宫颈细胞学涂片 ; 参考:《重庆大学》2014年博士论文


【摘要】:宫颈癌是严重威胁女性健康的恶性肿瘤之一,是发病率最高的三种癌症之一。据GLOBOCAN2012报告,全球2012年新增宫颈癌病例52.7万,约有26.5万妇女死于宫颈癌。中国2012年新增宫颈癌病例达6.2万,死亡病例达2.9万。宫颈癌有较长的病变期,,如果能在早期发现并及时治疗,治愈率较高,因而宫颈癌筛查对女性健康非常重要。 基于宫颈细胞学涂片的检查技术是目前医学上普遍采用的筛查技术,能够有效发现宫颈癌前病变和早期宫颈癌。然而基于人工判读的传统宫颈细胞学筛查方式存在工作量大、成本高、可靠性与准确性受到医师专业技术和主观情绪的影响等问题,因此开发基于计算机技术和图像处理技术的宫颈细胞学涂片自动判读系统对于宫颈癌的防治有着重要意义。 论文以宫颈细胞学涂片判读自动化为研究目标,将模拟细胞学医师判读宫颈细胞学涂片的方法作为自动判读的研究思路,结合宫颈细胞病理学知识,运用图像处理、本体建模、语义推理等技术,研究宫颈细胞学涂片自动判读的部分关键技术。论文的主要内容包括:宫颈细胞学涂片图像的粗分割、重叠细胞分割、单细胞精确分割;细胞图像的轮廓特征、染色质特征提取;细胞的图像特征、细胞学特征的本体建模与语义映射;宫颈细胞学涂片语义推理判读模型和相关判读规则构建。论文突破以分类器作为判读工具的传统方法,提出基于语义推理的宫颈细胞学涂片判读方法,为实现宫颈细胞学涂片判读自动化提出了一种新途径。论文的主要工作体现在以下5个方面: ①提出基于曲率的重叠细胞轮廓分离点检测方法和基于椭圆曲线拟合的重叠细胞分割方法。前者首先通过分析轮廓曲线曲率的正负值检测到凹区;然后根据曲率大小筛选出候选的重叠凹区,根据凹区宽度和凹区间距判定细胞重叠凹区;最后通过查找重叠凹区的曲率极点得到最终的细胞重叠接触点(即分离点)。后者首先将相邻分离点间的轮廓曲线上的点作为拟合数据,应用基于最小二乘法的椭圆拟合方法得到拟合椭圆,再过滤掉面积过大和过小的拟合椭圆,得到用以提取分离线的拟合椭圆;然后提取拟合椭圆上分离点间的弧线段作为分离线,据此分离重叠的细胞;最后分析细胞面积与细胞重叠区域面积的关系以确定分离是否有效。该方法能保持细胞的原有形态,同时降低了欠分离和过分离的概率。 ②提出基于极坐标系的梯度矢量流活动轮廓模型(PGVF Snake)。PGVFSnake首先把经过预处理的细胞图像从笛卡尔坐标系变换到极坐标系,计算基于极坐标系的边缘图;然后提出“浪沙抑制”算法优化PGVF模型中的边缘图,以消除细胞内部杂质对边缘图的干扰;最后使用边缘图作为活动轮廓模型中的外力,控制活动轮廓演化并收敛到细胞的真实边缘。与RGVF活动轮廓模型的对比实验结果表明,论文提出的方法在保证分割准确度的前提下,分割速度提高了五倍以上。 ③提出基于线性几何热流演化的细胞轮廓不规则度特征提取方法。该方法首先把细胞轮廓曲线进行几何热流演化,直到细胞轮廓演化为完全凸性为止,以此轮廓作为度量细胞轮廓不规则度的参考;然后比较演化后的轮廓曲线和原始轮廓曲线,提出不重叠区域总面积比、不重叠区域平均面积比等细胞轮廓不规则度特征描述子。在Herlev宫颈细胞图像数据集上的实验结果表明,论文所提方法提取的不规则度特征与宫颈细胞病变有明显的相关性。 ④提出基于数学形态学的染色质颗粒特征提取方法。该方法首先使用受限自适应直方图均衡方法增强细胞核图像的对比度;然后使用不同尺度的形态学基本结构元素进行开运算,计算累积尺度分布,通过对累积尺度分布求导,得到染色质颗粒尺度分布,并将颗粒尺度分布最大值所对应的尺度作为染色质颗粒特征描述子。在Herlev宫颈细胞图像数据集上的实验结果表明,论文所提方法提取的细胞染色质颗粒特征与宫颈细胞病变有明显的相关性。 ⑤基于本体与语义逻辑推理的宫颈细胞学涂片自动判读方法。该方法模拟细胞学医师判读宫颈细胞学涂片的过程,结合图像处理、知识表示、逻辑推理的相关理论,构建宫颈细胞学涂片推理判读模型。该方法首先建立与判读相关的细胞本体、图像特征本体、细胞学特征本体;然后提出基于语义规则的图像特征到语义特征的映射方法;最后建立基于本体与语义逻辑推理的宫颈细胞学涂片自动判读系统模型,并阐述了模型的原理。在Herlev宫颈细胞图像数据集上与k最邻近(kNN)分类器、支持向量机(SVM)的对比实验结果表明,论文提出的宫颈细胞学涂片判读方法对宫颈细胞病变有较高的灵敏度和特异度。
[Abstract]:Cervical cancer is one of the most malignant cancers that threaten women ' s health . It is one of the three cancers with the highest incidence . According to the report of GLOBOCAN2012 , the number of new cervical cancer increased by 52 . 7 million in 2012 , and about 265,000 women died of cervical cancer . In 2012 , the number of cervical cancer increased to 62,000 , with a mortality rate of 29,000 . Cervical cancer has a long lesion period . If it can be found in the early stage and treated in time , the cure rate is high , so cervical cancer screening is very important for women ' s health .

The examination technique based on cervical cytology smear is a commonly used screening technique in medicine , which can effectively detect cervical precancerous lesions and early cervical cancer . However , the traditional cervical cytology screening method based on manual interpretation has the problems of large workload , high cost , reliability and accuracy influenced by the professional technique and subjective emotion of the physician , etc . Therefore , it is important to develop cervical cytology smear automatic interpretation system based on computer technology and image processing technology for the prevention and treatment of cervical cancer .

Based on the study of cervical cytology smear , this paper studies the method of cervical cytology smear as the research thinking of automatic interpretation , combines the knowledge of cervical cell pathology , uses image processing , ontology modeling , semantic reasoning , and so on . The main contents of this paper include : coarse segmentation of smear image of cervical cytology , overlapping cell division , single cell precise segmentation ;
the contour feature of the cell image and the feature extraction of the chromatin ;
Image features , ontology modeling and semantic mapping of cellular features ;
Based on the traditional method of the classifier as the interpretation tool , this paper proposes a new method of cervical cytology smear based on semantic reasoning , which provides a new approach for the realization of cervical cytology smear automation . The main work of this paper is as follows :

( 1 ) the method of detecting the contour separation point of overlapped cells based on curvature and the overlapping cell segmentation method based on the elliptic curve fitting are proposed , the former firstly detects the concave area by analyzing the positive and negative values of the curvature of the contour curve ;
then selecting a candidate overlapping concave area according to the curvature size , and judging the cell overlapping concave area according to the concave area width and the concave area pitch ;
finally , obtaining the final cell overlapping contact point ( i.e . , the separation point ) by finding the pole of curvature of the overlapped concave area , and firstly , using the point on the contour curve between adjacent separation points as the fitting data , and applying the ellipse fitting method based on the least square method to obtain a fitting ellipse , and then filtering out the fitting ellipse with too large area and too small area to obtain a fitting ellipse for extracting the separation line ;
and then extracting the arc segment between the separated points on the fitting ellipse as a separation line , and separating the overlapping cells according to the separation line ;
Finally , the relationship between cell area and area of cell overlap is analyzed to determine whether the separation is effective . The method can preserve the original form of the cell , and reduce the probability of under - separation and over - separation .

Secondly , a gradient vector flow contour model ( PGVF Snake ) based on polar coordinate system is proposed . The PGVFSnake firstly transforms the preprocessed cell image from a Cartesian coordinate system to a polar coordinate system , and calculates an edge map based on a polar coordinate system ;
Then , an edge map in PGVF model is optimized by the " wave - sand suppression " algorithm to eliminate the interference of intra - cell impurities on the edge map ;
Finally , an edge map is used as the external force in the active contour model to control the evolution of the active contour and converge to the real edge of the cell . Compared with the RGVF active contour model , the results show that the proposed method improves the segmentation accuracy by more than five times under the premise of guaranteeing the segmentation accuracy .

( 3 ) The method of feature extraction of cell profile irregularity based on linear geometry heat flow evolution is proposed . The method includes the following steps : firstly , geometric thermal flow evolution is carried out on the contour curve of the cell until the evolution of the cell contour is complete convexity , and the contour is used as a reference for measuring the irregular degree of the contour of the cell ;
Then , the contour curve and the original contour curve after evolution are compared , and the feature descriptors of the area ratio of the non - overlapping area and the average area ratio of the non - overlapping area are presented . The experimental results on the Herlev cervical cell image data set show that the irregular characteristic extracted by the method is obviously related to the cervical cell lesion .

The method of feature extraction of chromatin particles based on mathematical morphology is proposed . Firstly , the contrast of nuclear image is enhanced by using constrained adaptive histogram equalization method .
Then using the basic structural elements of different scales to carry out the operation , the cumulative scale distribution is calculated , the scale distribution of the chromatin particles is obtained through the derivation of the cumulative scale distribution , and the scale corresponding to the maximum value of the particle size distribution is used as the characteristic description of the chromatin particles .

The invention relates to a cervical cytology smear automatic interpretation method based on ontology and semantic logic reasoning , which simulates the process of cervical cytology smear by a cytologic physician , combines image processing , knowledge representation and logical reasoning , and constructs a cervical cytology smear - based reasoning interpretation model .
then a mapping method based on semantic rules for image features to semantic features is proposed ;
Finally , a model of cervical cytology smear automatic interpretation system based on ontology and semantic logic reasoning is established , and the principle of the model is described . Compared with k nearest neighbor ( kNN ) classifier and support vector machine ( SVM ) on Herlev cervical cell image data set , the results show that cervical cytology smear interpretation method has high sensitivity and specificity for cervical cell pathological changes .

【学位授予单位】:重庆大学
【学位级别】:博士
【学位授予年份】:2014
【分类号】:R737.33;TP391.41

【参考文献】

相关期刊论文 前1条

1 卢智勇;梁光明;;基于多边形拟合的细胞图像特征提取算法[J];计算机仿真;2009年11期

相关博士学位论文 前3条

1 赵方辉;子宫颈癌筛查方法及策略的研究[D];北京协和医学院;2010年

2 杨峰;本体映射关键技术研究[D];吉林大学;2011年

3 范金坪;宫颈细胞图像分割和识别方法研究[D];暨南大学;2010年



本文编号:1804063

资料下载
论文发表

本文链接:https://www.wllwen.com/yixuelunwen/fuchankeerkelunwen/1804063.html


Copyright(c)文论论文网All Rights Reserved | 网站地图 |

版权申明:资料由用户14f37***提供,本站仅收录摘要或目录,作者需要删除请E-mail邮箱bigeng88@qq.com