Objective: To investigate a preoperative multi-sequence MRI-based radiomic nomogram for prediction of platinum-based chemotherapy sensitivity in patients with epithelial ovarian cancer (EOC). Methods: The complete data of 114 patients with EOC confirmed by surgery and pathology in Nantong Tumor Hospital of Nantong University from January 2015 to May 2020 were retrospectively analyzed, with an average age of 32-76 (57±8) years. All patients underwent platinum-based chemotherapy after maximal cytoreductive surgery. According to whether relapse occurred within 6 months, those patients were divided into platinum-resistant disease (PR, n=39) group and platinum-sensitive disease group (PS, n=75).All patients underwent MRI examination before treatment, and the 3-dimensional solid component of the tumor area of interest (ROI) on T2-weighted image (T2WI), diffusion weighted imaging (DWI) and T1-weighted image-enhanced image (T1CE) were manually delineated using Itk-snap software.Then AK software was imported for radiomics features extracting. They were randomly divided into training group (n=80) and validation group (n=34) in a ratio of 7∶3 (stratified sampling method). Firstly, the radiomics features were initially screened by the method of maximum correlation and minimum redundancy (mRMR), and features with the greatest predictive power were retained. Then, the LASSO regression analysis was performed to select the best features and construct the radiomics model. Univariate analysis was used to screen out clinical relevant factors, which combined with radiomic score (Radscore) was applied to develop a radiomics nomogram by multivariable logistic regression. Receiver operating characteristic (ROC) curve and decision curve analysis (DCA) were used to evaluate the predictive ability and clinical application value of radiomics model, clinical related factor model and radiomics nomogram. Results: Compared with the radiomics model (12 optimal radiomics features) and the clinical relevant factors model (residual disease, neutrophil count, carbohydrate antigen 199), the radiomics nomogram model demonstrated the best prediction performance: in the training groups, the AUC (Area Under the ROC Curve), accuracy, sensitivity, and specificity were 0.90 (95%CI:0.82-0.99), 90.0%, 89.0%, and 92.0%, respectively. In the validation groups, the AUC, accuracy, sensitivity, and specificity were 0.89 (95%CI:0.78-1.00), 85.0%, 87.0%, and 80.0%, respectively. DCA shows that the use of nomograms with a threshold in the range of 0.01 to 0.90 has a greater clinical application value in predicting the sensitivity of platinum chemotherapy in patients with EOC. Conclusion: The multi-sequence MRI-based radiomics nomogram has a high diagnostic value in predicting the sensitivity of platinum-based chemotherapy in patients with EOC.
目的： 构建基于多序列MRI影像组学列线图，并探究其用于预测上皮性卵巢癌（EOC）患者对铂类药物化疗敏感性的价值。 方法： 回顾性收集2015年1月至2020年5月南通大学附属肿瘤医院收治的114例经术后病理证实为EOC的患者资料，年龄32~76（57±8）岁。所有患者接受最大程度肿瘤细胞减灭术后均进行了铂类药物化疗，以化疗后6个月内是否复发，将患者分为铂耐药组（PR组）39例和铂敏感组（PS组）75例。所有患者治疗前均行MRI检查，在横轴位T2加权像（T2WI）、弥散加权成像（DWI）和T1加权像增强图像（T1CE）上沿着肿瘤实性成分轮廓勾画立体感兴趣区（3D ROI），应用AK软件提取影像组学特征。将患者按7∶3比例以分层抽样法随机分为训练集（80例）和验证集（34例），采用最大相关最小冗余（mRMR）方法对影像组学特征进行初筛，保留最大预测效能的特征，然后用最小绝对收缩选择算子（LASSO）回归分析进行特征降维，选择最优特征并构建影像组学模型。使用单因素分析筛选出临床相关因素，并结合影像组学评分（Radscore）使用多因素logistic回归分析构建影像组学列线图。应用受试者工作特征（ROC）曲线和决策曲线分析（DCA）评估影像组学模型、临床相关因素模型和影像组学列线图的预测能力和临床应用价值。 结果： 相较于影像组学（12个最优的影像组学特征）模型和临床相关因素（术后残留病灶、中性粒细胞计数、糖类抗原199）模型，影像组学列线图显示出最优的预测效能：在训练集中，ROC曲线下面积（AUC）、诊断准确率、灵敏度和特异度分别为0.90（95%CI：0.82~0.99）、90.0%、89.0%和92.0%；在验证集中，上述指标依次分别为0.89（95%CI：0.78~1.00）、85.0%、87.0%和80.0%。DCA显示阈值在0.01~0.90范围内使用影像组学列线图预测EOC患者铂类药物化疗敏感性的临床应用价值较大。 结论： 基于多序列MRI构建的影像组学列线图预测EOC患者对铂类药物化疗的敏感性具有较高的诊断效能。.