多参数MRI组学列线图模型预测子宫内膜样腺癌淋巴血管间隙侵犯  

Multi-parametric MRI radiomics-based nomogram model for predicting the lymphovascular space invasion of endometrial endometrioid adenocarcinoma

作　　者：马晓亮 沈敏花 马凤华 张国福 周建军 曾蒙苏 强金伟 MA Xiao-liang;SHEN Min-hua;MA Feng-hua;ZHANG Guo-fu;ZHOU Jian-jun;ZENG Meng-su;QIANG Jin-wei(Department of Radiology,Zhongshan Hospital,Fudan University,Shanghai 200032,China;Department of Radiology,Obstetrics and Gynecology Hospital,Fudan University,Shanghai 200090,China;Department of Radiology,Jinshan Hospital,Fudan University,Shanghai 200540,China)

机构地区：[1]复旦大学附属中山医院放射科,上海200032 [2]复旦大学附属妇产科医院放射科,上海200090 [3]复旦大学附属金山医院放射科,上海200540

出　　处：《复旦学报（医学版）》2024年第3期306-314,322,共10页Fudan University Journal of Medical Sciences

摘　　要：目的分析多参数磁共振成像(magnetic resonance imaging,MRI)组学列线图模型治疗前预测子宫内膜样腺癌(endometrial endometrioid adenocarcinoma,EEA)淋巴血管间隙侵犯(lymphovascular space invasion,LVSI)的可行性及价值。方法于2020年10月至2022年1月在复旦大学附属妇产科医院前瞻性收集205例EEA临床及MRI资料,按6∶4随机分为训练集(n=123)和验证集(n=82)。分别在T2加权成像、扩散加权成像(表观扩散系数图)及动态增强MRI序列勾画全肿瘤体积感兴趣区提取肿瘤影像组学特征。在训练集中,采用单变量分析及多变量Logistic回归分析筛选LVSI的独立预测因子,建立临床预测模型;采用最小绝对收缩和选择算法(least absolute shrinkage and selection operator,LASSO)进行特征筛选并建立影像组学标签;采用临床独立预测因子与组学标签构建临床-MRI组学列线图模型,并在验证集中进行模型验证。使用受试者操作特征曲线下面积(area under the receiver operating characteristic curve,AUC)评估模型效能,临床决策曲线评估模型临床应用价值。结果205例EEA中,LVSI(-)144例,LVSI(+)61例。患者绝经状态、CA125及CA199为LVSI(+)的临床独立预测因子,三者联合组成的临床预测模型AUC为0.714(训练集)和0.731(验证集)。从多参数MRI图像中共提取的8240个影像组学特征中筛选出5个最佳特征构建MRI组学标签,AUC为0.860(训练集)和0.759(验证集)。临床-MRI组学列线图模型AUC为0.887(训练集)和0.807(验证集),优于单独的临床模型及组学模型,且在较大的阈值概率范围内临床-MRI组学列线图模型可获得更大的临床净收益。结论基于多参数MRI组学的列线图模型可在治疗前有效预测EEA的LVSI状态,为临床管理决策提供有价值的参考,提高患者的临床获益。Objective To investigate the feasibility and value of a multi-parametric MRI radiomics-based nomogram model for pretreatment predicting the lymphovascular space invasion(LVSI)of endometrial endometrioid adenocarcinoma(EEA).Methods Preoperative MRI and baseline clinical characteristics of 205 EEA patients were prospectively collected from Oct 2020 to Jan 2022 in the Obstetrics and Gynecology Hospital,Fudan University,and randomly divided into training set(n=123)and validation set(n=82)in a 6∶4 ratio.The whole-tumor region of interest was manually drawn on T2-weighted imaging,diffusion-weighted imaging(apparent diffusion coefficient),and dynamic contrast-enhanced MRI,respectively,for radiomics features extraction.In the training set,univariate and multivariate Logistic regression analysis were used to select independent clinical predictors of LVSI(+)and construct the clinical model.The least absolute shrinkage and selection operator(LASSO)regression and multivariate Logistic regression analysis were used to select optimal radiomics features to form a radiomics signature.A combined nomogram model was established by integrating clinical independent predictors and the radiomics signature,and validated in the validation set.The predicting performance and clinical net benefit were evaluated by using the area under the receiver operating characteristic curve(AUC)and clinical decision curve analysis,respectively.Results Of the 205 EEA cases,144 cases were LVSI(-)and 61 cases were LVSI(+).Menopausal status,CA125,and CA199 were independent clinical predictors for the LVSI(+),and contributing to a clinical model with AUCs of 0.714(training)and 0.731(validation).From 8240 extracted radiomics features,five were selected to construct a MRI radiomics signature after de-redundancy and LASSO dimensionality reduction,yielding AUCs of 0.860(training)and 0.759(validation).The combined nomogram model showed AUCs of 0.887(training)and 0.807(validation),outperforming others and achieving maximum clinical benefit in a large range of th

关 键 词：影像组学 磁共振成像 子宫内膜癌 淋巴血管间隙侵犯 列线图 

分 类 号：R816.91[医药卫生—放射医学]