MRI纹理特征定量分析用于T2,3期直肠癌精准分期的初步研究  

Preliminary Study on Precise Staging of Rectal Cancer Stage T2,3 Based on Quantitative Analysis of MRI Texture Features

作　　者：王斌杰[1] 周彦汝 姜澳田 王长福[1] 李长波[1] 杨晓慧 张岚[2] WANG Bin-jie;ZHOU Yan-ru;JIANG Ao-tian;WANG Chang-fu;LI Chang-bo;YANG Xiao-hui;ZHANG Lan(Huaihe Hospital and Research Institute of Medical Image,Henan University,Kaifeng 475000,China;The Fist Affiliated Hospital of Henan University of Traditional Chinese Medicine,Zhengzhou 450000,China;School of Mathematics and Statistics,Henan University,Kaifeng 475000,China)

机构地区：[1]河南大学淮河医院河南大学医学影像研究所,河南开封475000 [2]河南中医药大学第一附属医院,河南郑州450000 [3]河南大学数学与统计学院,河南开封475000

出　　处：《中国肿瘤》2020年第7期554-560,共7页China Cancer

基　　金：2020年度河南省重点研发与推广专项(202102310087);河南省医学科技攻关计划(201404026)。

摘　　要：[目的]探讨磁共振成像(magnetic resonance imaging,MRI)纹理特征定量分析用于鉴别T2,3期直肠癌异质性及精准分期的价值。[方法]回顾性分析15例经手术病理证实T2及T3期病例,且在术前两周内行直肠MRI高分辨率平扫,每个病例选取显示病变较为满意T2WI轴位图像的若干层面,其中T2期8例选取41层,T3期7例选取40层,在MaZda软件中勾画病变感兴趣区(region of interest,ROI)提取病变纹理特征,用该软件提供的纹理特征选择方法中的交互信息(mutual information,MI)、Fisher系数(Fisher coefficient,Fisher)、分类错误概率联合平均相关系数(classification error probability combined with average correlation coefficients,POE+ACC)3种方法联合(Fisher+POE+ACC+MI,FPM)对提取纹理进行降维处理得到30个纹理特征,然后用该软件提供的纹理特征分类分析方法非线性分类分析(nonlinear discriminant analysis,NDA)对81个样本进行分类分析。采用组间与组内的平方和比例(between-category to within-category sums of squares,BW)和决策曲线分析法(decision curve analysis,DCA)筛选出3个最优特征,并比较两者差异。利用二元Logistic回归及ROC曲线计算三者独立及联合的诊断效能。[结果]81层图像分类的准确性为92.6%,以T2期图像作为阴性组、T3期图像作为阳性组计算敏感性为93.0%,特异性为93.2%,漏诊率为7.0%,误诊率为6.8%。以BW及DCA方法筛选出两组最优特征相同,前3个最优特征为S(4,0)平方和(sum of squares,SumOfSqs)、S(5,0)SumOfSqs、S(3,0)SumOfSqs,效能排序有差异,三者联合的敏感性和特异性分别为85.0%和70.4%,独立的敏感性和特异性分别为80.0%和62.9%,80.0%和62.9%,72.5%和60.3%。三者联合及独立的敏感性和特异性均低于30个纹理NDA分类结果。[结论]T2,3期直肠癌异质性有明显差异,MRI纹理特征定量分析能为T2,3期直肠癌的术前精准分期提供可靠客观依据。[Purpose]To investigate the quantitative analysis of MRI texture features in differentiating heterogeneity and precisely staging for rectal cancer stage T2,3.[Methods]Clinical and imaging data of15 patients with stage T2,3 rectal cancer confirmed by postoperative pathology,who underwent rectal high-resolution MRI scan two weeks before operation,were retrospectively analyzed.The images which were more satisfied with the axial T2 WI were selected,and 41 images were selected in 8 stage T2 cases and 40 images were selected in 7 stage T3 cases.By MaZda software,the region of interest(ROI)of the lesion was delineated for extracting the texture features.The mutual information(MI),Fisher coefficient(Fisher),and classification error probability combined with average correlation coefficients(POE+ACC)were provided by the software.The Fisher,POE+ACC and MI,and FPM(combination of Fisher,POE+ACC and MI)were used for screening texture features,then 30 texture features were obtained.Nonlinear classification analysis(NDA)was used for classifying 81 samples.Three optimal features were screened by the methods of BW(between-category to within-with-the-sums,BW)and DCA(decision curve analysis,DCA),then comparing the differences of the result.The diagnostic efficacy of three single features,and their combination was analyzed by using binary Logistic regression and ROC curve.[Results]The accuracy of image classification in 81 samples was 92.6%.The sensitivity of T2 image as negative group and T3 image as positive group was 93.0%,the specificity was 93.2%,the missed rate was 7.0%,and the misdiagnosis rate was 6.8%.The two sets of optimal characteristics were selected by BW and DCA methods,the first three optimal features were S(4,0)sum of squares(SumOfSqs),S(5,0)SumOfSqs,S(3,0)SumOfSqs,and the efficiency order was different.The sensitivity and specificity of the combination of three optimal features were 85.0%and 70.4%,and the sensitivity and specificity of three single features were 80.0%and 62.9%,80.0%and 62.9%,72.5%and 60.3%,respectiv

关 键 词：直肠癌 纹理特征分析 T分期 精准分期 磁共振成像 

分 类 号：R735.37[医药卫生—肿瘤]