肝细胞肝癌CT图像纹理分析分割技术的初步研究(1)
摘 要:目的 探讨机器自动分割肝癌CT图像纹理特征的可行性。方法 用强度、纹理、形状和边缘的图像特征来描述分割的情况,计算从操作者分割提取的特征与机器分割的相关性,测量不同操作者分割CT图像与机器分割CT图像的一致性。结果 操作者在选择不同层面时并不一致。操作者的分割结果也并非重叠。每个机器分割与其操作者手动分割的平均重叠程度与两个操作者之间的重叠程度相当(74% vs 69%)。机器分割与操作者手动分割组内相关性(ICC)结果表示纹理和强度特征是最显著的,边缘和形状特征最小。结论 本研究通过在每个操作者分割的最大圆来确定机器自动分割,从而有助于临床中可以更快、更准确的对CT图像进行分割。
关键词:影像组学;影像特征;分割
中图分类号:R735.7 文献标识码:A DOI:10.3969/j.issn.1006-1959.2018.11.027
文章编号:1006-1959(2018)11-0089-04
, http://www.100md.com
Preliminary Study on Segmentation Technique of Texture Analysis of Hepatocellular Carcinoma CT Image
WANG Qi-ling1,WANG Bing1,LI Yu-lin1,CHENG Guan-xun1,XIANG Zi-yun2
(1.Department of Medical Imaging,Shenzhen Hospital,Peking University,Shenzhen 518036,Guangdong,China;
2.Department of Imaging,Shenzhen Longgang District People's Hospital,Shenzhen 518172,Guangdong,China)
, 百拇医药
Abstract:Objective To explore the feasibility of automatic segmentation of liver cancer CT image texture features.Methods Intensity,texture,shape,and edge image features were used to describe the segmentation conditions.The correlation between the segmentation feature extracted from the operator and the machine segmentation was calculated,and the consistency between different operator segmented CT images and machine segmentation CT images was measured.Results Operators are inconsistent when choosing different levels.The operator's segmentation results are also not overlapping.The average degree of overlap between each machine segment and its operator's manual segmentation is comparable to the overlap between the two operators(74%vs69%).Machine segmentation and operator manual segmentation of intra-group correlation(ICC)results indicate that texture and intensity features are the most prominent,with minimal edge and shape features.Conclusion In this study,the automatic segmentation of the machine is determined by the maximum circle segmented by each operator,which helps to quickly and accurately segment the CT image in the clinic.
, 百拇医药
Key words:Radiomics;Image features;Segmentation
影像組学(radiomics)是指高通量地提取大量描述肿瘤特征性的影像特征,以深度挖掘图像信息,并从海量的影像图像中提取大量的定量特征,通过量化分析来提高诊断准确率[1]。提取定量特征的第一步是对肿瘤进行分割,对肿瘤CT图像的手动分割是一项非常耗时且繁杂的工作,需要在大量的CT图像中的各个层面进行肿瘤边界的手动分割,此外,手动分割CT图像并不能做到完全一致,分割的形状或位置具有可变性[2]。目前,已经开发了多种自动和半自动图像分割算法来勾画肿瘤边界,提高了一致性并减少了分割肿瘤所需的时间,但也并不总是能够区分肿瘤边界不清的病例[3],尤其在肝脏肿瘤中因边界难以确定,导致操作者在勾画肿瘤时不一致,最终造成诊断结果的不可靠。本研究尝试一种新的方法,可以用一个简单的过程自动分割图像,以获得肿瘤图像子集。通过在每个操作者分割的最大圆来确定机器自动分割,从而可以更快、更准确的对CT图像进行分割。
1资料与方法
1.1一般资料 回顾性分析2016年6月~2017年6月于北京大学深圳医院确诊为肝细胞癌的患者26例,其中男16例,女10例,年龄41~78岁,平均年龄(62.54±18.65)岁。, 百拇医药(王戚玲 汪兵 利玉林 成官迅 向子云)
关键词:影像组学;影像特征;分割
中图分类号:R735.7 文献标识码:A DOI:10.3969/j.issn.1006-1959.2018.11.027
文章编号:1006-1959(2018)11-0089-04
, http://www.100md.com
Preliminary Study on Segmentation Technique of Texture Analysis of Hepatocellular Carcinoma CT Image
WANG Qi-ling1,WANG Bing1,LI Yu-lin1,CHENG Guan-xun1,XIANG Zi-yun2
(1.Department of Medical Imaging,Shenzhen Hospital,Peking University,Shenzhen 518036,Guangdong,China;
2.Department of Imaging,Shenzhen Longgang District People's Hospital,Shenzhen 518172,Guangdong,China)
, 百拇医药
Abstract:Objective To explore the feasibility of automatic segmentation of liver cancer CT image texture features.Methods Intensity,texture,shape,and edge image features were used to describe the segmentation conditions.The correlation between the segmentation feature extracted from the operator and the machine segmentation was calculated,and the consistency between different operator segmented CT images and machine segmentation CT images was measured.Results Operators are inconsistent when choosing different levels.The operator's segmentation results are also not overlapping.The average degree of overlap between each machine segment and its operator's manual segmentation is comparable to the overlap between the two operators(74%vs69%).Machine segmentation and operator manual segmentation of intra-group correlation(ICC)results indicate that texture and intensity features are the most prominent,with minimal edge and shape features.Conclusion In this study,the automatic segmentation of the machine is determined by the maximum circle segmented by each operator,which helps to quickly and accurately segment the CT image in the clinic.
, 百拇医药
Key words:Radiomics;Image features;Segmentation
影像組学(radiomics)是指高通量地提取大量描述肿瘤特征性的影像特征,以深度挖掘图像信息,并从海量的影像图像中提取大量的定量特征,通过量化分析来提高诊断准确率[1]。提取定量特征的第一步是对肿瘤进行分割,对肿瘤CT图像的手动分割是一项非常耗时且繁杂的工作,需要在大量的CT图像中的各个层面进行肿瘤边界的手动分割,此外,手动分割CT图像并不能做到完全一致,分割的形状或位置具有可变性[2]。目前,已经开发了多种自动和半自动图像分割算法来勾画肿瘤边界,提高了一致性并减少了分割肿瘤所需的时间,但也并不总是能够区分肿瘤边界不清的病例[3],尤其在肝脏肿瘤中因边界难以确定,导致操作者在勾画肿瘤时不一致,最终造成诊断结果的不可靠。本研究尝试一种新的方法,可以用一个简单的过程自动分割图像,以获得肿瘤图像子集。通过在每个操作者分割的最大圆来确定机器自动分割,从而可以更快、更准确的对CT图像进行分割。
1资料与方法
1.1一般资料 回顾性分析2016年6月~2017年6月于北京大学深圳医院确诊为肝细胞癌的患者26例,其中男16例,女10例,年龄41~78岁,平均年龄(62.54±18.65)岁。, 百拇医药(王戚玲 汪兵 利玉林 成官迅 向子云)