摘要
目的探讨宫颈癌体素内不相干运动(IVIM)的直方图和纹理特征观察者内部及观察者间的可重复性。方法搜集2015年12月至2016年2月20例经本院1.5 T GE MR采集且病理证实为宫颈癌患者的IVIM图像,将这些图像导入GE AW4.6后处理工作站MADC软件中,分别生成标准表观扩散系数图(ADCstand)、快速表观扩散系数图(ADCfast)、慢速表观扩散系数图(ADCslow)、扩散分数(f)图;将这些生成的图像及b=1000 s/mm2的DWI图导入3D slicer软件中,由2名不同年资影像医师独立利用该软件在b=1000 s/mm2的DWI图上分别间隔一个月2次勾画病灶最大层面的感兴趣区(ROI),并将该ROI复制到各IVIM衍生的参数图上,再利用该软件计算肿瘤最大层面的常见直方图和纹理特征,共计11个特征,并使用t检验及组内相关系数(ICC)评价观察者内部、观察者之间测量特征的可重复性。结果在ADCstand图中,组内比较的10百分位数和组间比较的10百分位数、逆差矩为可重复特征;ADCfast图中组内比较的10百分位数、90百分位数、自相关性、对比度、能量、逆差矩、峰度、平均值、偏度、方差和组间比较的10百分位数、90百分位数、自相关性、对比度、能量、熵、逆差矩、峰度、方差为可重复特征;ADCslow图中组内比较的10百分位数、90百分位数、自相关性、平均值、方差和组间比较的10百分位数、自相关性、对比度、逆差矩、平均值、方差为可重复特征;f图中组内比较的自相关性、对比度、峰度、平均值、方差和组间比较的90百分位数、自相关性、对比度、熵、峰度、方差为可重复特征。结论宫颈癌IVIM衍生ADCfast、ADCslow、f图中提取的可重复特征的数量要稍高于ADCstand图;观察者内部认知水平的不同会对影像组学特征的可重复性产生影响。
Objective To investigate the intra-and inter-observer reproducibility of histogram and texture features derived from intravoxel incoherent motion(IVIM)diffusion-weighted MRI of cervical cancer.Methods The IVIM images of 20 patients with histologically proven cervical cancer from December 2015 to February 2016 acquired by 1.5 T GE MRI,were retrospectively collected.These images were imported into MADC software in the GE AW4.6 post-processing workstation to generate standard apparent diffusion coefficient(ADCstand),fast apparent diffusion coefficient(ADCfast)and slow apparent diffusion coefficient(ADCslow),diffusion fraction(f)maps,respectively.Then,these generated images and DWI images with b=1000 s/mm2 were imported into 3 D slicer software.Two radiologists with varying degrees of experience were blinded to segment the ROI of tumor’s largest transversal level two times apart for one month on DWI images with b=1000 s/mm2,and transplant the ROI to the parameter maps derived from each IVIM.The common histogram and texture features at the largest transversal cross-sections of tumors were calculated,totaling 11 features.The T test and intraclass correlation coefficient(ICC)were used to evaluate the reproducibility of intra-observer and inter-observer features.Results In ADCstand map,the 10 th percentile of intra-observer comparison and 10 th percentile,inverse difference moment(IDM)of inter-observer comparison are repeatable features.in ADCfast map,the 10 th percentile,90 th percentile,autocorrelation,contrast,energy,IDM,kurtosis,mean,skewness,variance of intra-observer comparison and 10 th percentile,90 th percentile,autocorrelation,contrast,energy,entropy,IDM,kurtosis and variance of inter-observer comparison are repeatable features.In ADCslow map,10 th percentile,90 th percentile,autocorrelation,mean,variance of intra-observer comparison and 10 th percentile,autocorrelation,contrast,IDM,mean,variance of inter-observer comparison are repeatable features.In f map,autocorrelation,contrast,kurtosis,mean,variance of intra-observer comparison and 90 th percentile,autocorrelation,contrast,entropy,kurtosis,variance of inter-observer comparison are repeatable features.Conclusion The number of repeatable features extracted from ADCfast,ADCslow and f maps of cervical cancer is slightly higher than that from ADCstand maps.Thus,observers with different experiences have an impact on the reproducibility of histogram and texture features.
作者
陈浩
袁子龙
郑丽丽
贺瑶瑶
彭伟
陈迢
张照喜
CHEN Hao;YUAN Zilong;ZHENG Lili(Department of Radiology,Hubei Provincial Cancer Hospital,Wuhan,Hubei Province 430079,P.R.China)
出处
《临床放射学杂志》
CSCD
北大核心
2020年第8期1577-1582,共6页
Journal of Clinical Radiology
基金
科技部国家重点专项计划项目(编号:2016YFC0103400)
关键词
宫颈癌
体素内不相干运动
直方图
纹理特征
可重复性
Cervical cancer
Intravoxel incoherent motion
Histogram
Texture features
Reproducibility
