摘要
目的探讨基于CT的影像组学模型对腺性膀胱炎(CG)与膀胱癌的鉴别诊断效能。方法回顾性分析经病理证实的40例CG(CG组)和70例膀胱癌(膀胱癌组),均于术前接受盆腔平扫和三期增强CT扫描。分别在平扫、动脉期、静脉期及延迟期CT图像上勾画三维容积感兴趣区(VOI),提取影像组学特征。将按7∶3随机将患者分入训练集与测试集。对各期CT影像组学特征数据行归一化处理,以最小冗余最大相关法(mRMR)、最小绝对收缩和选择算法(LASSO)及5折交叉验证进行特征降维;采用多因素Logistic回归分析建立影像组学模型。以受试者工作特征(ROC)曲线评价各模型的诊断效能,采用Delong检验评价其效能差异,以决策曲线分析评估模型临床应用价值。采用Hosmer-Lemeshow检验和校正曲线评估模型拟合度。结果分别基于平扫、动脉期、静脉期、延迟期影像组学特征建立了模型1、2、3、4,各纳入4、7、5及6个特征;其鉴别诊断CG与膀胱癌的曲线下面积(AUC)均>0.80,且Delong检验表明不同模型间AUC值差异无统计学意义(P均>0.05)。模型2在测试组的AUC=0.939,高于其他模型。决策曲线分析表明,模型2用于临床鉴别CG与膀胱癌的净获益最高,Hosmer Lemeshow拟合优度检验显示模型2预测结果与实际结果差异无统计学意义(训练集:χ^(2)=8.75,P=0.36;测试集:χ^(2)=4.72,P=0.79);校正曲线显示模型2的拟合效果较好。结论基于各时相盆腔CT影像组学模型均可用于辅助临床鉴别诊断CG与膀胱癌,其中基于动脉期影像组学模型的诊断效能最高。
Objective To explore the differential diagnostic efficacy of based on CT radiomics model for cystitis glandularis(CG)and bladder cancer.Methods Data of 40 patients with CG(CG group)and 70 with bladder cancer(bladder cancer group)confirmed by pathology were retrospectively analyzed.All patients received pelvic plain and three-phase enhanced CT scanning before operation.The volume of interest(VOI)were delineated on plain CT as well as arterial phase,venous phase and delayed phase of enhanced CT images,respectively,and the radiomics features were extracted.The patients were randomly divided into training set and test set according to the ratio of 7∶3.After normalizing these data including CT radiomics characteristics of all 4 phases,the minimum redundancy maximum correlation method(mRMR),the least absolute shrinkage and selection operator(LASSO)and 5-fold cross validation were used to reduce the dimension of features.Multivariate Logistic regression analysis was used to develop models.The receiver operating characteristic(ROC)curve was used to evaluate the diagnostic efficacy of each model,and Delong test was used to evaluate the difference of efficacies among 4 models.Decision curve analysis was performed to evaluate the clinical utility value of models.The Hosmer-Lemeshow test and calibration curve were performed to assess the goodness-of-fit of the models.Results Four models were established,namely model 1,2,3 and 4,including 4,7,5 and 6 features,based on the radiomics features of plain CT,arterial phase,venous phase and delayed phase enhance CT images,respectively.The area under the curve(AUC)values of 4 models for differential diagnosis of CG and bladder cancer were all larger than 0.80,and Delong test showed that there was no significant difference of AUC among all models(all P>0.05).AUC of model 2 in test set was 0.939,higher than that of the other models.Decision curve analysis showed that model 2 had the highest net benefit for clinical differentiation of CG and bladder cancer.The Hosmer-Lemeshow goodness
作者
陈鹏飞
俞咏梅
吴琦
陈基明
CHEN Pengfei;YU Yongmei;WU Qi;CHEN Jiming(Department of Medical Imaging Center, Yijishan Hospital of Wannan Medical College, Wuhu 241001, China)
出处
《中国介入影像与治疗学》
北大核心
2021年第6期360-365,共6页
Chinese Journal of Interventional Imaging and Therapy
