| 刘肖敏,邹煜,俞静婧,汪小琛,林昱含,秦佳乐.临床-MRI影像组学联合模型鉴别交界性卵巢肿瘤与上皮性卵巢癌[J].中国医学影像技术,2025,41(10):1701~1705 |
| 临床-MRI影像组学联合模型鉴别交界性卵巢肿瘤与上皮性卵巢癌 |
| Clinical-MRI radiomics combined model for differentiating borderline ovarian tumor from epithelial ovarian cancer |
| 投稿时间：2025-05-09 修订日期：2025-10-05 |
| DOI：10.13929/j.issn.1003-3289.2025.10.021 |
| 中文关键词: 卵巢肿瘤 诊断,鉴别 磁共振成像 影像组学 |
| 英文关键词:ovarian neoplasms diagnosis, differential magnetic resonance imaging radiomics |
| 基金项目:浙江省"尖兵领雁＋X"研发攻关计划(2024C03184)、国家卫生健康委科学研究基金——浙江省卫生健康重大科技计划重大项目(WKJ-ZJ-2408)。 |
|
| 摘要点击次数: 151 |
| 全文下载次数: 52 |
| 中文摘要: |
| 目的 观察临床-MRI影像组学联合模型鉴别交界性卵巢肿瘤(BOT)与上皮性卵巢癌(EOC)的价值。方法 回顾性纳入经术后病理确诊且术前接受盆腔MR检查的139例BOT(BOT组)及307例EOC(EOC组)患者,按7∶3比例随机划分训练集(n=312)与测试集(n=134)。以多因素logistic回归筛选BOT与EOC的临床独立预测因子并构建临床模型。分别于T2WI、弥散加权成像(DWI)及表观弥散系数(ADC)图中病灶感兴趣容积(V OI)提取影像组学特征,基于训练集以极限梯度提升(XGBoost)构建单及多序列MRI影像组学模型,根据测试集受试者工作特征曲线的曲线下面积(AUC)选取其中最优者,结合临床独立预测因子建立临床-MRI影像组学联合模型。利用测试集比较临床模型、最优MRI影像组学模型及联合模型鉴别BOT与EOC的效能,以沙普利加性解释(SHAP)解析综合最优模型的关键预测特征。结果 患者年龄、糖类抗原153(CA153)及糖类抗原125(CA125)均为BOT与EOC的独立预测因子(P均＜0.05)。多序列MRI影像组学模型为最优MRI影像组学模型;以之联合临床独立预测因子构建的联合模型鉴别测试集BOT与EOC的效能(AUC=0.929)高于临床模型(AUC=0.881)及多序列MRI影像组学模型(AUC=0.871)(P均＜0.05)。SHAP蜂群图显示,联合模型中,特征重要性排序前10者包括年龄、CA153及CA125,以及ADC与DWI的熵、峰度及灰度级非均匀性等。结论 基于多序列MRI影像组学联合临床特征构建的临床-MRI影像组学联合模型可有效鉴别BOT与EOC。 |
| 英文摘要: |
| Objective To explore the value of clinical-MRI radiomics combined model for differentiating borderline ovarian tumor (BOT) from epithelial ovarian cancer (EOC). Methods Totally 139 patients with BOT (BOT group) and 307 patients with EOC (EOC group) confirmed by postoperative pathology and underwent preoperative pelvic MRI were retrospectively enrolled and randomly divided into training set (n=312) and test set (n=134) at a ratio of 7∶3. Multivariable logistic regression was used to identify independent clinical predictors for differentiating BOT and EOC, then a clinical model was constructed. Radiomics features were extracted from the volumes of interest (VOI) of lesions on T2WI, diffusion weighted imaging (DWI) and apparent diffusion coefficient (ADC) images, respectively, and single-sequence and multi-sequence MRI radiomics models were built using extreme gradient boosting (XGBoost) based on data in training set. The optimal MRI radiomics model was selected according to the highest area under the curve (AUC) in test set, and a clinical-MRI radiomics combined model was constructed combined the optimal radiomics model with independent clinical predictors. The performances of clinical model, the optimal MRI radiomics model and the combined model for differentiating BOT and EOC were compared in test set.SHapley Additive exPlanations (SHAP) analysis was applied to interpret key predictive features in the best model. Results Patients’ age, carbohydrate antigen 153 (CA153) and carbohydrate antigen 125 (CA125) were all independent predictors for differentiating BOT and EOC (all P＜0.05). Multi-sequence MRI radiomics model was the optimal MRI radiomics model. The combined model showed superior performance (AUC=0.929) for discriminating BOT and EOC compared with clinical model (AUC=0.881) and multi-sequence MRI radiomics model (AUC=0.871) (both P＜0.05). SHAP beeswarm plot revealed that the top 10 important features of combined model included age, CA153 and CA125, as well as entropy, kurtosis and gray level non-uniformity from ADC and DWI sequences. Conclusion Clinical-MRI radiomics combined model based on multi-sequence MRI radiomics features and clinical features could be used to effectively differentiate BOT from EOC. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
