磁共振影像组学对直肠癌新辅助治疗后病理完全反应的评估价值  

The value of MRI radiomics for evaluation of pathological complete response of rectal cancer after neoadjuvant treatment

作　　者：朱洁[1] 沈浮[1] 袁渊[1] 王敏杰[1] 白辰光[2] 王颢[3] 邵成伟[1] ZHU Jie;SHEN Fu;YUAN Yuan(Department of Radiology,Changhai Hospital of Shanghai,Shanghai 200433,China)

机构地区：[1]上海长海医院影像医学科,上海200433 [2]上海长海医院病理科,上海200433 [3]上海长海医院肛肠外科,上海200433

出　　处：《放射学实践》2022年第4期426-431,共6页Radiologic Practice

摘　　要：目的:探讨基于MR T_(2)WI的影像组学方法对直肠癌接受新辅助治疗(nCRT)后病理完全反应(pCR)状态的评估价值。方法:回顾性分析2019年1月-2020年12月在我院接受新辅助放化疗(nCRT)后行手术切除的99例局部进展期直肠癌(locally-advanced rectal cancer,LARC)患者的病例资料。根据术后病理检查结果,分为pCR组(22例)及非pCR组(77例)。在高分辨率T_(2)WI上勾画病灶的容积感兴趣区(volume of interest,VOI)并提取其组学特征,采用最小绝对收缩和选择算子(LASSO)算法进行特征降维,筛选出与pCR相关的最佳组学特征。将所有病例按照7∶3的比例随机分为两组:训练集(68例)和测试集(31例),建立支持向量机(support vector machine,SVM)机器学习模型,绘制其ROC曲线,并计算曲线下面积(AUC)、敏感度和特异度。结果:共提取1409个组学特征,经降维后得到11个最有价值的组学特征。建立的SVM机器学习模型在测试集中预测pCR的AUC为0.798(95%CI:0.615~0.920),符合率为83.87%,敏感度为85.71%,特异度为83.33%。结论:基于高分辨T_(2)WI的影像组学特征有助于预测nCRT后直肠癌pCR状态,可指导临床决策。Objective:To explore the value of radiomics based on high resolution T_(2)WI in evaluating pathologic complete response(pCR)of rectal cancer after neoadjuvant therapy.Methods:This retrospective study included 99 patients with locally advanced rectal cancer(LARC)who underwent excision after neoadjuvant chemoradiotherapy(nCRT)from January 2019 to December 2020 in our hospital.According to postoperative pathological examination,patients were divided into pCR group(n=22)and non-pCR group(n=77).After manually delineating the volume of interest(VOI)of each lesion on the rectal high-resolution T_(2)WI images,the radiomics features were extracted.The least absolute shrinkage and selection operator(LASSO)algorithm was used to select the optimal features which were statistically related to pCR.All samples were randomly divided into training set(n=68)and test set(n=31)for machine learning according to a ratio of 7∶3.A support vector machine(SVM)model was established,and receiver operating characteristic curve(ROC)was drawn for eva-luating its prediction efficacy.The area under the curve(AUC),sensitivity and specificity were calculated.Results:Totally 1409 radiomics features were extracted from the VOI on T_(2)W images.11 optimal features were selected out by LASSO method.The AUC of the established SVM model for predicting pCR in test set was 0.798(95%confidence interval:0.615~0.920),the accuracy was 83.87%,the sensitivity was 85.71%,and the specificity was 83.33%.Conclusion:The radiomics features based on high-resolution T_(2)WI images are helpful for evaluating pCR state of rectal cancer after neoadjuvant treatment,and are beneficial to personalized treatment strategy.

关 键 词：直肠癌 新辅助治疗 影像组学 磁共振成像 机器学习 

分 类 号：R735.4[医药卫生—肿瘤] R445.2[医药卫生—临床医学]