多参数MRI影像组学模型术前对早期子宫内膜癌的风险预测

Multi-parameter MRI Radiomics in Preoperative Prediction of Risk Stratification in Early-stage Endometrial Carcinoma

目的:分析基于多参数MRI影像组学模型在术前预测低风险子宫内膜癌(EC)的可行性及临床价值。方法:回顾性收集204例国际妇产科联盟(FIGO)分期I期和Ⅱ期的EC患者资料,分为训练集(中心1,n=87)和测试集(中心2,n=117)。所有患者均于术前2周内行盆腔MR平扫及增强扫描。分析比较低风险与非低风险患者的临床特征差异,分别在轴位脂肪抑制T2加权成像(FS-T2WI)、表观弥散系数(ADC)和对比增强T1加权成像(CE-T1WI)图像上逐层勾画全肿瘤感兴趣区,提取与风险因素高度相关的影像组学特征并筛选建立影像组学模型,采用受试者工作特征曲线对模型效能进行评估。结果:低风险组和非低风险组组间患者年龄、肿瘤体积差异均有统计学意义(P<0.05),但多变量逻辑回归分析显示它们并非独立风险因素。单参数FS-T2WI、ADC和CE-T1WI影像组学模型的AUC值在训练集分别为0.730、0.724和0.757,在测试集分别为0.750、0.743和0.746。3个参数的融合模型在训练集和测试集的AUC值分别为0.827(95%CI 0.742~0.912)和0.823(95%CI0.748~0.897),优于单参数模型(所有P<0.05)。结论:多参数MRI影像组学模型对早期低风险EC有较好的预测价值,并具有可重复性。

Purpose:To analyze the value of radiomics based on multi-parameter MRI in preoperative prediction of low-risk early-stage endometrial carcinoma(EC).Methods:A total of 204 patients with Federation International of Gynecology and Obstetrics(FIGO)stage I and II EC were divided into training set(from center 1,n=87)and testing set(from center 2,n=117).Patient's clinical characteristics were compared between low risk group and non-low risk group.The whole tumor was delineated on axial fat suppression T2-weighted imaging(FS-T2WI),apparent diffusion coefficient(ADC)map and contrast-enhanced T1-weighted imaging(CE-TIWI),and the radiomics features highly correlated with risk factors were extracted and selected to establish the radiomics models.Receiver operating characteristic curve was used to evaluate the performance of the models.Results:There were significant differences in age and tumor volume between low risk group and non-low risk group(all P<0.05).However,multivariate logistic regression analysis showed that they were not the independent risk factors.The area under curves(AUCs)of the single-parameter of FS-T2WI,ADC and CE-T1WI radiomics models were 0.730,0.724,0.757 in the training set,and 0.827(95%CI 0.742-0.912)in the training set and 0.823(95%CI 0.748-0.897)in the testing set,which were superior to those of single-parameter models.Conclusion:Multi-parameter MRI radiomics model could be a helpful tool for preoperative prediction of low-risk early-stage EC.