影像组学为基础的列线图模型对子宫内膜癌淋巴血管间隙浸润的诊断价值  

The Predictive Value of Radiomics Based Nomogram Model for Lymphatic Vascular Space Infiltration of Endometrial Carcinoma

作　　者：崔玉杰[1] 郑涛[1] 杨林沙 石清磊 刘兰祥[1] 刘德丰[1] CUI Yujie;ZHENG Tao;YANG Linsha;SHI Qinglei;LIU Lanxiang;LIU Defeng(Medical Imaging Center,First Hospital of Qinhuangdao,Qinhuangdao Hebei 066000,China;Magnetic Resonance Products Division,Siemens LTD,Beijing 100000,China)

机构地区：[1]秦皇岛市第一医院医学影像中心,河北秦皇岛066000 [2]西门子有限公司磁共振产品事业部,北京100000

出　　处：《中国医疗设备》2021年第5期85-90,共6页China Medical Devices

基　　金：国家重点研发计划重点专项(2017YFB1002300)。

摘　　要：目的以表观扩散系数加权图像(Apparent Diffusion Coefficient,ADC)为基础,建立影像组学列线图模型,并对子宫内膜癌淋巴血管间隙浸润(Lymph Vascular Space Invasion,LVSI)情况进行预测。方法回顾性研究225名术后病理证实的子宫内膜癌(Endometrial Cancer,EC)患者资料,其中LVSI(+)93例。从ADC图中提取组学参数后应用LASSO回归进行筛选,并计算组学评分(Radscore)。应用Logistic回归建立组学列线图(Nomogram)模型(ModelN),与年龄、CA125和肿瘤体积组成的临床模型(ModelC)进行比较。应用校准曲线(Calibration Curve)对ModelC拟合优度进行检验。结果ModelC预测LVSI的能力有限(训练集AUC:0.760,95%CI:0.675~0.823;验证集AUC:0.825,95%CI:0.718~0.904)。共筛选出3个与LVSI关系最为密切的组学特征。联合Radscore后所建立的ModelN,在训练集和验证集中都体现出了较好的预测LVSI的能力(训练集AUC:0.907,95%CI:0.851~0.940;验证集AUC:0.922,95%CI:0.827~0.964)。与ModelC相比,ModelN预测能力提升(Delong test,训练集P<0.001,测试集P=0.035),校准曲线显示出了ModelN具有良好的拟合优度(Hosmer–Lemeshow test,训练集P=0.102,测试集P=0.287)。结论ADC为基础影像组学模型能够预测EC的LVSI。Objective Based on the Apparent Diffusion Coefficient(ADC),establish an imaging nomogram model and predict the Lymph Vascular Space Invasion,LVSI of endometrial cancer.Methods This retrospective study included 225 Endometrial Cancer,EC patients confirmed by postoperative pathology,including 93 cases of LVSI(+).After extracting radiomics parameters from ADC,LASSO regression was applied for parameters selection and then the radscore was calculated.Logistic regression was used to establish a nomogram model(ModelN)for comparison with a clinical model(ModelC)consisting of age,CA125,and tumor volume.Calibration Curve was used to test the goodness of fit of ModelC.Results Using ModelC to predict the LVSI(+)of EC was limited in both the cohort area under the curve(AUC:0.760;95% confidence interval(CI):0.675~0.823) and test cohorts(AUC:0.825;95%CI:0.718~0.904).A total of three omics features were screened out that were most closely related to LVSI.ModelN,established based on the combination of RadScore,showed the better level of predictive ability in both the training(AUC:0.90795%CI:0.851~0.940)and test cohorts(AUC:0.922;95%CI:0.827~0.964).In comparison with ModelC,the discrimination ability of ModelN showed improvement(Delong test,training cohorts:P<0.0001 and test cohorts:P=0.035).Calibration curves suggested a good fit for probability(Hosmer-Lemeshow test,P=0.102 and P=0.287 for the training and test cohorts,respectively).Conclusion ADC-based radiomics model can predict LVSI of EC preoperatively.

关 键 词：影像组学 列线图 子宫内膜癌 淋巴血管间隙浸润 表观扩散系数 

分 类 号：R737.33[医药卫生—肿瘤]