基于T2加权MRI瘤周和瘤内影像组学模型术前预测上皮性卵巢癌FIGO分期

Preoperative prediction of FIGO stage of epithelial ovarian cancer based on T2-weighted MRI peritumoral and intratumoral radiomics models

目的探讨基于T2加权MRI瘤周和瘤内影像组学模型预测上皮性卵巢癌(epithelial ovarian cancer,EOC)国际妇产科联盟(International Federation of Gynecology and Obstetrics,FIGO)分期的准确性与价值。材料与方法回顾性分析苏州市立医院(中心一)及苏州大学附属第一医院(中心二)共189例EOC患者的临床及影像资料,其中87例为FIGOⅠ/Ⅱ期,102例为FIGOⅢ/Ⅳ期。中心一的患者数据用于模型训练,中心二的数据作为外部验证集。基于肿瘤边界勾画感兴趣区域(region of interest,ROI),并分别向外扩展2 mm、4 mm、6 mm、8 mm和10 mm以获得多重瘤周信息。从瘤内及各瘤周ROI分别提取1223个影像组学特征。使用单因素分析、相关性分析、最小冗余最大相关及最小绝对收缩和选择算子(least absolute shrinkage and selection operator,LASSO)算法进行特征筛选。比较各瘤周外扩范围的影像组学模型在训练集中的效能,以选择最优的外扩范围构建瘤周模型;并进一步构建瘤内模型及临床模型。通过列线图将三者结合构建联合模型。将各模型应用于外部验证集,并通过受试者工作特征曲线(receiver operating characteristic,ROC)分析评估诊断效能。采用DeLong检验对比模型之间诊断效能的差异。结果瘤周影像组学模型在2 mm的外扩范围下表现最优,其训练集ROC曲线下面积(area under the curve,AUC)达0.840。在外部验证集中,联合模型诊断性能最佳,其具有最佳的准确度(74.2%)、特异度(80.8%)及AUC(0.837)。基于DeLong检验,联合模型显著优于瘤周模型(P=0.047)。结论基于T2加权MRI的瘤周、瘤内影像组学方法,有潜力有效预测上皮性卵巢癌的FIGO分期,其中联合瘤周瘤内影像组学及临床信息的模型表现最佳。该模型有望帮助临床医生更好地评估患者状况并制订个性化的治疗方案。

Objective:To investigate the accuracy and value of peritumoral and intratumoral radiomics models based on T2-weighted MRI in predicting the International Federation of Gynecology and Obstetrics(FIGO)stage of epithelial ovarian cancer(EOC).Materials and Methods:A total of 189 EOC patients from Suzhou Municipal Hospital(Center I)and the First Affiliated Hospital of Soochow University(CenterⅡ)were retrospectively collected,including 87 patients with FIGO stage I/Ⅱand 102 patients with FIGO stageⅢ/Ⅳ.The data from Center I were used for model training,while the data from CenterⅡwere used as an external validation set.The region of interest(ROI)was drawn based on the tumor boundary and extended outwardly by 2 mm,4 mm,6 mm,8 mm,and 10 mm to obtain multiple peritumoral information.A total of 1223 radiomics features were extracted from both intra-and peritumoral regions.Univariate analysis,correlation analysis,minimum redundancy maximum relevance,and least absolute shrinkage and selection operator(LASSO)algorithm were employed for feature selection.The performance of peritumoral radiomics models with different peritumoral extension ranges was compared in the training set to determine the optimal extension for constructing the peritumoral model.Subsequently,both intratumoral and clinical models were developed.The combined model was constructed based on intratumoral,peritumoral,and clinical features using a nomogram.Each individual model was subsequently applied to the external validation set,and their diagnostic performance was assessed through receiver operating characteristic(ROC)analysis.The DeLong test was employed to compare the differences in diagnostic efficacy between these models.Results:The peritumoral radiomics model demonstrated superior performance within an extended range of 2 mm,exhibiting an area under the ROC curve(AUC)of 0.840 in the training set.In the external validation set,the combined model exhibited optimal diagnostic performance,showcasing exceptional accuracy(74.2%),specificity(80.8%),and AUC(0.837).According to the DeLong test,the combined model significantly outperformed the peritumoral model(P=0.047).Conclusions:The T2-weighted MRI-based peritumoral and intratumoral radiomics method demonstrated promising potential in effectively predicting the FIGO stage of EOC.Notably,the model combining peritumoral,intratumoral radiomics,and clinical information exhibits superior performance.This advanced model is anticipated to assist clinicians in accurately assessing patient conditions and devising personalized treatment plans.