摘要
目的基于超声组学特征与临床参数构建联合模型,观察其预测高级别浆液性卵巢癌(HGSOC)淋巴结转移(LNM)效能。方法回顾性纳入401例HGSOC,根据入院时间分为训练集(n=322,LNM亚组138例、无LNM亚组184例)与测试集(n=79,LNM亚组35例、无LNM亚组44例)。收集术前临床参数纳入机器学习筛选器,筛选最优临床参数组合以构建临床模型。于术前末次超声检查显示病灶实性成分最大切面灰阶图、CDFI及与之正交的最大切面灰阶图中勾画病灶ROI并提取其超声组学特征,筛选对于预测LNM最有意义的特征,构建超声组学模型并计算影像组学评分(Radscore);基于最优临床参数及Radscore构建联合模型。绘制采用受试者工作特征(ROC)曲线,计算曲线下面积(AUC),评估各模型预测效能。结果糖类抗原(CA125)和大网膜增厚状况是预测LNM的最优临床参数组合;以之联合Radscore构建的联合模型在训练集和测试集的AUC分别为0.930和0.881,其在训练集的预测效能优于临床模型(P=0.002)及超声组学模型(P<0.001),在测试集优于临床模型(P=0.010)而与超声组学模型差异无统计学意义(P=0.285)。结论基于超声组学特征与临床参数的联合模型预测HGSOC LNM效能较好。
•Objective To construct a combined model based on ultrasound radiomics and clinical parameters,and to explore its efficiency for predicting lymph node metastasis(LNM)of high-grade serous ovarian cancer(HGSOC).Methods A total of 401 HGSOC patients were retrospectively enrolled and divided into training set(n=322,LNM subgroup 138 and non LNM subgroup 184)and testing set(n=79,LNM subgroup 35 and non LNM subgroup 44)according to hospital admission time.Preoperative clinical parameters were collected and subjected to machine learning-based feature selection for optimal clinical features,which were used to construct a clinical model.Then grayscale image and CDFI planes showing the largest crosssection of the solid component of lesion were selected along with the orthogonal maximum cross-sectional grayscale plane from the last preoperative ultrasound examination.ROI was delineated,radiomics features were extracted,and the most effective features for predicting LNM were selected to construct a ultrasound radiomics model,and the radiomics score(Radscore)was calculated.Finally a combined model was constructed based on the optimal clinical parameters and Radscore.Receiver operating characteristic(ROC)curves were drawn,the area under the curve(AUC)was calculated to evaluate the performance of these models.Results Carbohydrate antigen(CA125)and omental thickening status were the optimal clinical features combination for predicting LNM,and were used to construct the combined model together with the Radscore.AUC of the combined model was 0.930 in training set and 0.881 in testing set.In training set,the combined model demonstrated superior predictive performance compared to clinical model(P=0.002)and ultrasound radiomics model(P<0.001).In testing set,the combined model performed better than clinical model(P=0.010),but being not significantly different with ultrasound radiomics model(P=0.285).Conclusion Combined model based on ultrasound radiomics features and clinical parameters had good predictive efficacy for HGSOC LNM.
作者
齐玥
庄连婷
张宇晴
黄瑛
QI Yue;ZHUANG Lianting;ZHANG Yuqing;HUANG Ying(Department of Ultrasound,Shengjing Hospital of China Medical University,Shenyang 110004,China)
出处
《中国医学影像技术》
2025年第11期1795-1799,共5页
Chinese Journal of Medical Imaging Technology
关键词
卵巢上皮性癌
淋巴转移
超声检查
影像组学
carcinoma,ovarian epithelial
lymphatic metastasis
ultrasonography
radiomics
