摘要
目的探讨基于T2WI的3D纹理分析评估宫颈癌组织学分级的价值。方法回顾性分析经病理证实的175例宫颈癌患者,其中高分化41例(高分化组),中分化76例(中分化组),低分化58例(低分化组),术前均接受常规MR平扫及增强扫查。采用ITK-SNAP软件勾画感兴趣体积(VOI),以LIFEx软件计算获取41个纹理参数;比较3组间纹理参数差异,以组间差异有统计学意义的纹理参数构建Logistic回归模型,评价其评估宫颈癌组织学分级的效能。结果低、中、高分化组间,区域灰度不均匀度(GLNUz)、区长度不均匀度(ZLNU)、能量(GLCM-Energy)、强度(Busyness)、游程灰度级不均匀度(GLNUr)、游程长度不均匀度(RLNU)、体积(Volume-vx)及容积(Volume-ml)8个参数差异有统计学意义(P均<0.05)。低分化组与高分化组间8个纹理参数差异均有统计学意义(P均<0.05),中分化与高分化组间Energy、GLNUz、ZLNU差异有统计学意义(P均<0.05)。低、中、高分化组间差异有统计学意义的8个纹理参数均与组织学分级相关(|r|=0.491~0.567)。低分化与高分化组间8个差异有统计学意义纹理参数鉴别二者的AUC为0.711~0.774,以其构建的Logistic回归模型的AUC为0.875,敏感度87.50%,特异度77.78%。中分化与高分化组间3个差异有统计学意义的纹理参数的AUC为0.685~0.717,以此构建的Logistic回归模型的AUC为0.753,敏感度78.75%,特异度72.92%。结论基于T2WI的3D纹理分析对术前预测宫颈癌组织学分级有一定价值,其模型诊断效能更高。
Objective To explore the value of 3D texture analysis(TA)based on T2WI for predicting histological grade of cervical cancer.Methods Data of 175 patients of cervical cancer confirmed by pathology were retrospectively analyzed,including 41 cases of high differentiation(high differentiation group),76 of middle differentiation(middle differentiation group)and 58 cases of low differentiation(low differentiation group)cervical cancer.All patients underwent conventional plain and enhanced MR scanning before operation.The volume of interest(VOI)was delineated by using ITK-SNAP software,and 41 texture parameters were calculated and obtained with LIFEx software.Then the texture parameters were compared among 3 groups.Taken texture parameters statistically different among groups,Logistic regression models for predicting histological grade of cervical cancer before surgical operations were established,and their effectiveness were analyzed.Results Statistical significant differences of 8 parameters(GLNUz,ZLNU,GLCM-Energy,Busyness,GLNUr,RLNU,Volume-vx and Volume-ml)were found among 3 groups(all P<0.05).There were statistically significant differences of 8 texture parameters between low differentiation group and high differentiation group(all P<0.05),while tatistical significant differences of Energy,GLNUz and ZLNU were detected between middle differentiation group and high differentiation group(all P<0.05).Eight texture parameters being statistically different among low,middle and high differentiation groups were all correlated with histological grade(|r|=0.491-0.567).AUC value of 8 statistically different texture parameters between low and high differentiation groups were 0.711-0.774,of Logistic regression model based on these parameters was 0.875,with sensitivity of 87.50%and specificity of 77.78%.AUC of 3 texture parameters being statistically different between middle and high differentiation groups were 0.685-0.717,of Logistic regression model was 0.753,with sensitivity of 78.75%and specificity of 72.92%,respectively.Concl
作者
尹进学
卢斌贵
杨佩瑜
钟熹
陈志军
桂思
洪璇阳
李颖慧
孙紫情
李建生
YIN Jinxue;LU Bingui;YANG Peiyu;ZHONG Xi;CHEN Zhijun;GUI Si;HONG Xuanyang;LI Yinghui;SUN Ziqing;LI Jiansheng(Department of Radiology,Affiliated Cancer Hospital and Institute of Guangzhou Medical University,Guangzhou 510095,China)
出处
《中国医学影像技术》
CSCD
北大核心
2021年第1期86-90,共5页
Chinese Journal of Medical Imaging Technology
关键词
子宫颈肿瘤
磁共振成像
纹理分析
组织学分级
uterine cervical neoplasms
magnetic resonance imaging
texture analysis
histological grade
