摘要
目的:探讨基于矢状位T2WI灰度共生矩阵(GLCM)对Ⅰ期子宫内膜癌(EC)肌层浸润深度的评估价值。方法:回顾性分析本院经手术后病理证明的47例EC病例(包含Ⅰ_(a)期即浅肌层浸润,25例;Ⅰ_(b)期即深肌层浸润,22例)。用软件ITK-SNAP于矢状位T2WI序列勾画病变多层面感兴趣区域,再经过软件AK(Artificial Intelligence Kit,V3.2.0R版本)图像预处理后得到标准化图像,并采集灰度共生矩阵纹理特征48个(包括8类参数:集群突出、集群阴影、相关、能量、熵、Haralick相关、惯量、逆差矩;6个角度计算:全角、全角SD、0°、45°、90°、135°;步长为1)。Ⅰ_(a)期组与Ⅰ_(b)期组纹理特征比较用独立样本t检验(呈正态分布且方差齐)或Mann-Whitney U检验(呈非正态分布或者方差不齐),组内多参数降维行因子分析。以受试者操作特征(ROC)曲线分析计算降维后纹理参数曲线下面积、界值、敏感度、特异度。结果:3类参数(共13个纹理特征)差异有统计学意义(P<0.05),即相关(全角、0°、45°、90°),惯量(全角、全角SD、0°、45°、90°),逆差矩(全角、0°、45°、90°)。而集群突出、集群阴影、能量、熵、Haralick相关差异无统计学意义(P>0.05)。13个纹理参数降维成2种主成分,其中相关(Correlation_angle45_offset1)和惯量(Inertia_AllDirection_offset1)得分分别最高(分别为0.928、0.904)。Correlation_angle45_offset1判断Ⅰ_(b)期EC的ROC曲线下面积为0.756,界值为2.07×10^(-3),敏感度77.3%,特异度72.0%;Inertia_AllDirection_offset1判断Ⅰ_(a)期EC的ROC曲线下面积为0.718,界值为165.89,敏感度68.0%,特异度63.8%。结论:基于矢状位T2WI序列灰度共生矩阵可用于评估Ⅰ期EC肌层浸润深度,进而为临床对其治疗方法的择取及预后判断提供帮助,并展示良好的应用前景。
Objective:To evaluate the value of sagittal T2WI MRI based on gray level co-occurrence matrix(GLCM)in the diagnosis of the depth of myometrial invasion in stage Ⅰ endometrial cancer(EC).Methods:A retrospective analysis of 47 cases of endometrial cancer confirmed by surgery and pathology in our hospital were retrospectively analysed(Ⅰ_(a) stage,superficial myometrial invasion,accounting for 25 cases.Ⅰ_(b) stage,deep myometrial invasion,accounting for 22 cases).The multi-faceted region of interest was delineated by the software ITK-SNAP in the sagittal T2WI sequence,and then subjected to image preprocessing by software AK(Artificial Intelligence Kit,V3.2.0R version).Standardized images were obtained and 48 numbers of gray level co-occurrence matrix texture features were collected(including 8 types of parameters:Cluster prominence,cluster shade,correlation,GLCM energy,GLCM entropy,Haralick correlation,inertia,inverse difference moment.6 angle calculations:All direction,all direction SD,0°,45°,90°,135°.Step size was 1).The comparison of texture features between group Ⅰ_(a) and group Ⅰ_(b) was performed by independent sample t test(normal distribution and homogeneity variance)or Mann-Whitney U test(non-normal distribution or heterogeneity variance),and then factor reduction was used in the group to reduce dimensionalities.The receiver operating characteristic(ROC)curve was used to calculate the area under the curve,threshold,sensitivity and specificity of the texture parameter curve after dimension reduction.Results:The difference of the three types of parameters(13 texture features)was statistically significant(P<0.05),ie correlation(all direction,0°,45°,90°),inertia(all direction,all direction SD,0°,45°,90°),inverse difference moment(all direction,0°,45°,90°).The cluster prominence,cluster shade,GLCM energy,GLCM entropy,Haralick correlation were not statistically significant(P>0.05).The 13 texture parameters were reduced to two principal components,and the scores of correlation(Correlation_angle45_offset1)and inertia(Inertia_AllDirection_offset1)were the highest(0.928,0.904,respectively).Correlation_angle45_offset1 determined that the area under the ROC curve of theⅠ_(b) stage EC was 0.756.The threshold was 2.07×10^(-3).The sensitivity was 77.3%,and the specificity was 72.0%.Inertia_AllDirection_offset1 determined that the area under the ROC curve of the stageⅠ_(a) EC was 0.718.The threshold was 165.89.The sensitivity was 68.0%,and the specificity was 63.8%.Conclusion:Based on the gray level co-occurrence matrix of the sagittal T2WI sequence,the depth of myometrial invasion of stage I endometrial cancer can be evaluated,and then it helps the clinical selection of its treatment methods and prognosis judgment,which has showed a good application prospect.
作者
姚易明
叶靖
高慧
郭晓丽
王丽娟
吴晶涛
YAO Yiming;YE Jing;GAO Hui;GUO Xiaoli;WANG Lijuan;WU Jingtao(Dalian Medical University,Liaoning Dalian 116044,China;Department of Radiology,Northern Jiangsu People's Hospital,Jiangsu Yangzhou 225001,China)
出处
《现代肿瘤医学》
CAS
北大核心
2021年第8期1404-1409,共6页
Journal of Modern Oncology
基金
扬州市科教强卫重点人才资助项目(编号:ZDRC201873)。
关键词
子宫内膜癌
分期
灰度共生矩阵
纹理分析
因子分析
endometrial cancer
staging
gray level co-occurrence matrix
texture analysis
factor analysis
