基于不同机器算法的MRI影像组学-临床联合模型诊断非酒精性脂肪性肝病的对比研究  

A Comparative Study of Different MRI Radiomics-Clinical Combined Modeling in Diagnosis of Nonalcoholic Fatty Liver Disease

作　　者：邓丹丹 李曼 李兵[1] 杜勇[1] DENG Dandan;LI Man;LI Bing;DU Yong(Department of Radiology,Affiliated Hospital of North Sichuan Medical College,Nanchong 637000,China;Department of Research and Development,Shanghai United Imaging Intelligence Co.,Ltd.)

机构地区：[1]川北医学院附属医院放射科,四川省南充市637000 [2]上海联影智能有限公司研发部

出　　处：《中国医学计算机成像杂志》2024年第6期683-692,共10页Chinese Computed Medical Imaging

基　　金：南充市市校合作科研专项(NSMC20170301)。

摘　　要：目的:比较基于不同机器算法的MRI影像组学联合临床因素构建的模型诊断非酒精性脂肪性肝病(NAFLD)的效能,以找到最佳诊断模型。方法:回顾性收集于川北医学院附属医院就诊的207例拟诊NAFLD并接受MRI检查者资料,其中包括89例NAFLD患者和118例非NAFLD对照者,按8∶2比例随机分为训练集(n=165,71例NAFLD、94例非NAFLD对照)和测试集(n=42,18例NAFLD、24例非NAFLD对照)。通过单因素及多因素logistic回归分析筛选脂肪肝的临床独立预测因子。基于MRI中同相位(OP)-反相位(IP)序列图像提取影像组学特征并采用最小绝对收缩和选择算子(LASSO)选择最优影像组学特征。运用6种机器算法联合临床特征及影像组学特征,建立6种影像组学联合诊断模型。运用受试者工作特征曲线(ROC)评估各模型诊断效能,并计算ROC曲线下面积(AUC)、灵敏度、特异度、准确度、精确度和F1评分,运用校准曲线评估各模型预测概率与实际观测结果的一致性,运用决策曲线分析(DCA)评价各模型的临床价值。结果:甲胎蛋白(AFP)、极低密度脂蛋白-胆固醇(VLDL-C)和精神病分别为NAFLD的临床独立预测因子。基于6种机器算法的MRI影像组学-临床联合模型中,除基于高斯过程及支持向量机算法的模型的诊断效能较差,基于其余4种机器算法(决策树、随机森林、逻辑回归及梯度提升)的模型诊断NAFLD的效能均较好,但梯度提升机器算法影像组学联合模型的整体诊断效能最高,其AUC、灵敏度、特异度、准确率、精确度、F1分数在训练组中分别为0.999、1.000、1.000、0.999、0.999、1.000,在验证组中分别为0.999、0.994、1.000、0.976、1.000、0.971。校准曲线结果示梯度提升机器算法影像组学联合模型的临床效能最高的预测概率和实际观测值的一致性最好。DCA的结果显示,训练组和验证组中,梯度提升机器算法影像组学联合模型的临床效能最高。结论:Purpose:To study the efficacy of different machine algorithms for MRI radiomics combined with clinical factors for the diagnosis of nonalcoholic fatty liver disease(NAFLD),and to find the best model.Methods:The data from 207 adults who underwent MRI examination for suspected NAFLD at the Affiliated Hospital of North Medical College were retrospectively collected,including 89 patients with NAFLD and 118 non-NAFLD controls.They were randomly divided into a training set(n=165,71 cases of NAFLD,94 non-NAFLD controls)and a test set(n=42,18 cases of NALFD,24 non-NAFLD controls)in an 8∶2 ratio.Clinical independent predictors of fatty liver were screened by univariate and multivariate logistic regression analysis.Radiomics features were extracted based on the in phase and out of phase(OP-IP)sequence images in MRI and the optimal radiomics features were selected using the least absolute shrinkage and selection operator(LASSO)algorithm.Six machine algorithms were used to establish six sets of radiomics diagnosting models that combined the clinical features and the radiomics features.The diagnostic efficacy of each model was evaluated using the receiver operating characteristic(ROC)curve.The consistency of the predicted probabilities of each model using calibration curves with actual observations,and the clinical value of each model was evaluated using decision curve analysis(DCA).Results:Alpha-fetoprotein(AFP),very low density lipoprotein cholesterol(VLDL-C)and psychosis were clinical independent predictors of NAFLD.In the MRI radiomics-clinical combined models,except for the Gaussian process-and SVM machine algorithms-based models,the diagnostic performance of the other four models(base on decision tree,random forest,logistic and XGBOOST machine algorithms,respectively)were all well in diagnosing NAFLD.The radiomics combined model based on XGBOOST machine algorithm had the highest diagnostic efficacy for diagnosing NAFLD,with the AUC,sensitivity,specificity,accuracy,precision,and F1 scores of 0.999,1.000,1.000,0.999,0.9

关 键 词：影像组学 机器算法 非酒精性脂肪性肝病 磁共振成像 诊断 

分 类 号：R445.2[医药卫生—影像医学与核医学]