摘要
目的观察静脉期CT影像组学联合临床特征预测上皮性卵巢癌(EOC)患者乳腺癌易感基因(BRCA)突变的价值。方法回顾性纳入111例经手术病理及BRCA检测确诊的EOC患者,按8∶2比例划分训练集[n=90,35例BRCA突变(+)及55例BRCA突变(-)]与测试集[n=21,8例BRCA突变(+)及13例BRCA突变(-)]。采用单及多因素logistic回归(LR)分析临床及CT资料,筛选EOC患者BRCA突变的独立预测因素并建立临床模型。基于静脉期CT提取及筛选病灶最佳影像组学特征,计算影像组学评分(Radscore),并分别以随机森林(RF)、支持向量机(SVM)和LR建立机器学习(ML)模型并选择其中最优者,基于Radscore联合独立预测因素构建联合模型。评估各模型预测效能及临床价值。结果人附睾蛋白4为EOC患者BRCA突变的独立预测因素,以之构建的临床模型在训练集和测试集的曲线下面积(AUC)分别为0.648、0.742。RF、SVM、LR模型在训练集的AUC分别为0.726、0.763、0.860,在测试集分别为0.708、0.750、0.700,以SVM模型为最优ML模型。联合模型在训练集和测试集的AUC分别为0.819、0.783,其在训练集的AUC高于临床模型(P=0.022);其余模型两两比较差异均无统计学意义(P均>0.05)。决策曲线分析显示,当阈值大于0.15时,联合模型的临床价值高于临床及SVM模型。结论静脉期CT影像组学联合临床特征能有效预测EOC患者BRCA突变。
Objective To observe the value of venous phase CT radiomics combined with clinical features for predicting breast cancer susceptibility gene(BRCA)mutation in patients with epithelial ovarian cancer(EOC).Methods A total of 111 EOC patients diagnosed by surgical pathology and BRCA detection were retrospectively enrolled and divided into training set(n=90,35 BRCA mutations[+]and 55 BRCA mutations[-])and test set(n=21,8 BRCA mutations[+]and 13 BRCA mutations[-])at the ratio of 8∶2.Clinical and CT data were analyzed using univariate and multivariate logistic regression(LR)to screen independent predictors of BRCA mutations in EOC patients,and then a clinical model was established.Based on venous phase CT,the best radiomics features of EOC lesions were extracted and screened,radiomics score(Radscore)was calculated.Machine learning(ML)models were established using random forest(RF),support vector machine(SVM)and LR,respectively,and the optimal ML model was screened.Finally a combined model was constructed based on Radscore and independent predictors.The predictive efficacy and clinical value of each model were evaluated.Results Human epididymis protein 4 was the independent predictor of BRCA mutation in EOC patients,and the area under the curve(AUC)of clinical model was 0.648 and 0.742 in training and test sets,respectively.AUC of RF,SVM and LR model was 0.726,0.763 and 0.860 in training set,0.708,0.750 and 0.700 in test set,respectively,and SVM model was the optimal ML model.AUC of combined model was 0.819 and 0.783 in training and test set,respectively,which in training set was higher than that of clinical model(P=0.022).No significant difference of AUC was found by pairwise comparison of other models in both training and test set(all P>0.05).Decision curve analysis showed that when the threshold was larger than 0.15,the clinical value of combined model was higher than that of clinical and SVM models.Conclusion Venous phase CT radiomics combined with clinical features could effectively predict BRCA mutation in EOC patients.
作者
徐梦莉
赵艳萍
马焱
马尔克亚·卡马力拜克
李莉
XU Mengli;ZHAO Yanping;MA Yan;MAERKEYA·Kamalibaike;LI Li(Department of Gynaecology,Affiliated Tumor Hospital of Xinjiang Medical University,Urumqi 830011,China;Department of Nuclear Medicine,Affiliated Tumor Hospital of Xinjiang Medical University,Urumqi 830011,China)
出处
《中国医学影像技术》
北大核心
2025年第6期952-957,共6页
Chinese Journal of Medical Imaging Technology
基金
“天山英才”科技创新领军人才项目(2022TSYCLJ0033)。
