摘要
目的:比较传统影像特征与多序列影像组学模型在鉴别诊断Ⅰ型与Ⅱ型上皮性卵巢癌(EOC)中的效能,并验证传统模型与影像组学模型相结合能否提升预测EOC分型的效能。方法:搜集2015年1月-2019年6月经我院手术病理证实为EOC,且术前行MRI检查的61例患者,共80个病灶(Ⅰ型30个,Ⅱ型50个)。建立传统模型,包括常规MRI形态学特征及ADC值,通过多因素logistic回归筛选有统计学意义的特征;建立包括FS-T 2WI、DWI和T 1增强图像的影像组学模型,每个序列提取1070个影像组学特征,采用单变量分析和最小绝对收缩选择算法(LASSO)筛选重要特征,最后将传统模型与多序列组学模型相结合以建立混合模型。通过ROC曲线分析、校准曲线和决策曲线(DCA)分析验证各模型的预测性能。采用Delong检验比较不同模型之间的AUC值差异。结果:传统模型显示出最高的性能,AUC为0.95(95%CI:0.90~0.99),混合模型的AUC为0.96(95%CI:0.9~1.0),混合模型的性能与传统模型差异无统计学意义(P>0.05)。校准曲线分析结果表明,传统模型具有最高的可靠性。分层分析结果显示了影像组学模型在早期区分两种肿瘤类型中的潜力。结论:传统模型可成为区分I型与II型EOC的有效工具,影像组学模型有可能在早期更好地区分EOC类型。
Objective:To compare the diagnostic performance of the traditional imaging features and the multi-sequence Radiomics model in the differential diagnosis of typeⅠand typeⅡof epithelial ovarian cancer(EOC),and to verify whether the combination of the traditional model and the Radiomics model can improve the predictive efficacy of the histological types of EOC.Methods:A total of 61 patients with EOC confirmed by surgical pathology were collected in our hospital from January 2015 to June 2019,and underwent MRI examination before operation.80 lesions in total(30 lesions of typeⅠand 50 lesions of typeⅡ)were detected.The features including the morphological characteristics and ADC value of conventional MRI were screened by multivariate logistic regression,and the traditional model was established.A radiomics model was established in addition,including FS-T 2WI,DWI and T 1 enhanced images.1070 radiomics features were extracted from each sequence,and univariate analysis and least absolute shrinkage selection operator(LASSO)were used to screen important features.Finally,the mixed model was established by combining the traditional model with the multi-sequence Radiomics model.The prediction performance of each model was verified by ROC curve analysis,calibration curve analysis and decision curve analysis(DCA).Delong test was used to compare the differences of AUC values among different models.Results:The traditional model showed the highest performance with an AUC of 0.95(95%CI:0.90~0.99)and the mixed model with an AUC of 0.96(95%CI:0.9~1.0).There was no statistical difference in the performance between the mixed model and the traditional model(P>0.05).The calibration curve showed that the traditional model has the highest reliability.Hierarchical analysis showed the potential of Radiomics model in distinguishing between the two tumor types at an early stage.Conclusion:The traditional model can be an effective tool for differentiating typeⅠfrom typeⅡof EOC,and the Radiomics model may better distinguish EOC types
作者
钱洛丹
吴慧
牛广明
任嘉梁
崔艳芬
蔚纳
QIAN Luo-dan;WU Hui;NIU Guang-ming(Department of Radiology,Affiliated Hospital of Inner Mongolia Medical University,Hohhot 010050,China)
出处
《放射学实践》
CSCD
北大核心
2021年第5期621-627,共7页
Radiologic Practice
基金
内蒙古自然科学基金资助项目(2017MS(LM)0837)。
关键词
上皮性卵巢癌
卵巢肿瘤
组织学分型
影像组学
磁共振成像
Epithelial ovarian cancer
Ovarian Cancer
Histological classification
Radiomics
Magnetic resonance imaging
