多模态MRI影像组学联合临床危险因素术前预测无淋巴结转移的直肠癌脉管浸润  

Preoperative prediction of vascular invasion in rectal cancer patients without lymph node metastasis based on multimodal MRI imaging features combined with clinical risk factors

作　　者：杨燕 魏焕焕 付芳芳 魏巍[1,2] 吴亚平 吉祥[3] 王梅云 YANG Yan;WEI Huanhuan;FU Fangfang;WEI Wei;WU Yaping;JI Xiang;WANG Meiyun(Department of Medical Imaging,the People's Hospital of Zhengzhou University,Zhengzhou 450003,China;Henan Provincial People's Hospital,Henan Key Laboratory of Neurological Imaging,Zhengzhou 450003,China;School of Computers and Artificial Intelligence,Zhengzhou University,Zhengzhou 450001,China)

机构地区：[1]郑州大学人民医院医学影像科,郑州450003 [2]河南省人民医院河南省神经疾病影像诊断与研究重点实验室,郑州450003 [3]郑州大学计算机与人工智能学院,郑州450001

出　　处：《磁共振成像》2023年第1期94-99,110,共7页Chinese Journal of Magnetic Resonance Imaging

基　　金：河南省科技攻关项目(编号:212102310689);河南省医学科技攻关计划联合共建项目(编号:LHGJ20210001、LHGJ20210005)。

摘　　要：目的探讨基于轴位脂肪抑制T2加权成像(fat suppression-T2 weighted imaging,FS-T2WI)及T1加权对比增强(T1-weighted contrast-enhanced,T1CE)序列联合临床预测因子创建的临床-影像组学模型在预测无淋巴结转移的直肠癌患者术前淋巴血管浸润(lymphovascular invasion,LVI)中的应用价值。材料与方法回顾性纳入2016年12月至2021年12月河南省人民医院收治的行MRI扫描且术后病理证实为直肠癌的221例患者的病例及影像资料。采用单因素和多因素logistic回归分析LVI阳性组和LVI阴性组的临床数据,以确定LVI的独立预测因子;通过ITK-SNAP软件手动勾画肿瘤全层感兴趣区(region of interest,ROI),利用开源软件PyRadiomics提取影像组学特征,采用SPSS随机数字表法按8∶2比例将患者分为训练集(177例)和测试集(44例),并经过特征降维构建影像组学标签;基于临床预测因子是否加入影像组学模型构建了四个预测模型。根据受试者工作特征曲线下面积(area under the curve,AUC)、敏感度和特异度评估不同预测模型的诊断效能。结果肿瘤最大直径是直肠癌患者LVI的独立临床预测因子(P<0.05)。单独的FS-T2WI、T1CE序列以及两者联合(FS-T2WI+T1CE)的测试集AUC分别为0.757、0.802及0.869,FS-T2WI+T1CE联合临床预测因子的临床-影像组学模型诊断效能最好,测试集AUC为0.898(95%CI:0.769、0.968)。结论本研究构建的临床-影像组学模型具有较高的诊断效能,可以辅助临床对无淋巴结转移的直肠癌患者术前个体化LVI预测,改善治疗方案。Objective:To explore the application value of the clinical-radiomics model based on axial fat suppression T2-weighted imaging(FS-T2WI)and T1-weighted contrast-enhanced(T1CE)sequences combined with clinical predictors in the prediction of preoperative lymphovascular invasion(LVI)in patients with rectal cancer without lymph node metastasis.Materials and Methods:The cases and imaging data of 221 patients with rectal cancer who underwent MRI scan and were confirmed by postoperative pathology in Henan Provincial People’s Hospital from December 2016 to December 2021 were retrospectively included.Univariate and multivariate logistic regression were used to analyze the clinical data of the LVI positive group and the LVI negative group to determine the independent predictors of LVI.The full-layer region of interest(ROI)of tumor was manually delineated by ITK-SNAP software,and the open source software Py Radiomics was used to extract the radiomics features.Patients were divide into the training set(177 cases)and the test set(44 cases)according to the ratio of 8∶2 by SPSS random number table method,and the radiomics signature was constructed after feature dimension reduction.Four prediction models were constructed based on whether clinical predictors were included in the image omics model.The diagnostic efficacy of different prediction models was evaluated according to the area under curve(AUC)of receiver operating characteristic(ROC),sensitivity and specificity.Results:Maximum tumor diameter was independent predictors of LVI in patients with rectal cancer(P<0.05).The AUC of FS-T2WI,T1CE and their combination(FS-T2WI+T1CE)was 0.757,0.802 and 0.869,respectively.The FS-T2WI+T1CE combined with clinical predictors clinical-radiomics model had the best diagnostic performance,with an AUC of 0.898(95%CI:0.769,0.968)in the test set.Conclusions:The clinical-radiomics model constructed in this study has a high diagnostic efficiency,which can assist the clinical prediction of preoperative individualized LVI in rectal cancer patien

关 键 词：直肠癌 淋巴脉管浸润 影像组学 磁共振成像 logistic回归 

分 类 号：R445.2[医药卫生—影像医学与核医学] R735.37[医药卫生—诊断学]