基于MRI及US影像组学-临床联合模型诊断非酒精性脂肪性肝病  

Diagnosis of non-alcoholic fatty liver disease based on MRI and US radiomics-clinical combined modeling

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 North of Medical College,Nanchong 637000,China;Department of Research and Development,Shanghai United Imaging Intelligence Co.,Ltd.,Shanghai 200232,China)

机构地区：[1]川北医学院附属医院放射科,四川南充637000 [2]上海联影智能有限公司研发部,上海200232

出　　处：《分子影像学杂志》2024年第5期474-483,共10页Journal of Molecular Imaging

基　　金：南充市市校合作科研专项(NSMC20170301)。

摘　　要：目的探讨基于MRI及US影像组学-临床联合模型诊断非酒精性脂肪性肝病(NAFLD)的价值。方法回顾性分析2021年3月~2023年9月川北医学院附属医院207例患者的MRI及US图像,包括89例NAFLD患者和118例非NAFLD患者,按8:2比例随机分为训练集(n=165,NAFLD 71例、非NAFLD 94例)和测试集(n=42,NAFLD18例、非NAFLD 24例)。通过单因素及多因素Logistic回归分析筛选NAFLD的独立临床预测因子,建立临床模型。基于MRI中OP-IP序列及US图像提取并筛选影像组学特征,建立影像组学模型,分别为MRI影像组学模型、US影像组学模型及MRI-US影像组学联合模型。联合临床特征及影像组学特征,分别建立MRI影像组学-临床联合模型、US影像组学-临床联合模型、MRI-US影像组学-临床联合模型;采用受试者ROC曲线评估各模型诊断NAFLD的效能,运用决策曲线分析评价各模型的临床价值。结果甲胎蛋白、极低密度脂蛋白胆固醇和精神病分别为NAFLD的临床独立预测因子,用于建立临床模型。基于MRI及US图像分别筛选出19、16个最佳影像组学特征,用于建立MRI影像组学模型、US影像组学模型及MRI-US影像组学联合模型。联合临床特征及影像组学最佳特征,分别建立MRI影像组学-临床联合模型、US影像组学-临床联合模型、MRI-US影像组学-临床联合模型。训练集中,临床模型、MRI影像组学模型、US影像组学模型、MRI-US影像组学联合模型、MRI影像组学-临床联合模型、US影像组学-临床联合模型、MRIUS影像组学-临床联合模型诊断NAFLD阳性的AUC分别为0.99、1.00、0.99、1.00、1.00、0.99、1.00,在测试集中分别为0.94、1.00、0.89、0.99、1.00、0.98、1.00。决策曲线分析结果显示训练集和测试集中,MRI-US影像组学-临床联合模型的效能最高。结论基于MRI及US影像组学-临床联合模型对NAFLD具有较高的诊断价值。Objective To investigate the value of diagnosing non-alcoholic fatty liver disease(NAFLD)based on MRI and US radiomics-clinical combined model.Methods The MRI and US images of 207 patients,including 89 NAFLD patients and 118 non-NAFLD patients from the Affiliated Hospital North of Medical College from March 2021 to September 2023 were retrospectively analyzed and randomly divided into a training set(n=165,71 cases of NAFLD and 94 cases of non-NAFLD)and a test set(n=42,18 cases of NAFLD,24 cases of non-NAFLD).Independent clinical predictors of NAFLD were screened by univariae ad multivariate Logistic regression analysis.Then clinical models were established.Based on the OP-IP sequence in MRI and US images,we extracted and screened the radiomics features and established the radiomics models,which were MRI radiomics model,US radiomics model and MRI-US radiomics combined model.The MRI radiomics-clinical combined model,US radiomics-clinical combined model and MRI-US radiomics-clinical combined model were established by combining the clinical features and radiomics features;the efficacy of each model in the diagnosis of NALFD was evaluated by using the ROC curves of the subjects,and the clinical value of each model was evaluated by using the decision curve analysis.Results AFP,VLDL-C and mental disease were clinically independent predictors of NAFLD.At the same time,clinical model was established.Based on MRI,ninteen radiomics features were screened out,and based on US images,sixteen radiomics features were screened out.Meanwhile,the radiomics features of filtering out were used to establish MRI radiomics model,US radiomics model and MRI-US radiomics combined model.Clinical features and the best radiomics features were combined,which were used to build MRI radiomics-clinical combined model,US radiomics-clinical combined model,and MRI-US radiomics-clinical combined model.In the training set,the AUC of the clinical model,the MRI radiomics model,the US radiomics model,the MRI-US radiomics combined model,the MRI radiomics-

关 键 词：影像组学 深度学习 非酒精性脂肪性肝病 磁共振成像 超声成像 诊断 

分 类 号：R575.5[医药卫生—消化系统] R445.2[医药卫生—内科学]