基于图像人工分割建立乙肝肝硬化门脉压力预测模型  

Prediction model of portal pressure for HBV-related cirrhosis based on radiomics feature

作　　者：谭钦天 周锟 艾英杰 姜思雨 于泽宽 吴玲 黄晓铨 陈世耀 Tan Qintian;Zhou Kun;Ai Yingjie;Jiang Siyu;Yu Zekuan;Wu Ling;Huang Xiaoquan;Chen Shiyao(Department of Gastroenterology and Hepatology,Zhongshan Hospital,Fudan University,Shanghai 200032,China;Academy for Engineering&Technology,Fudan University,Shanghai 200433,China)

机构地区：[1]复旦大学附属中山医院消化科,上海200032 [2]复旦大学工程与应用技术研究院,上海200433

出　　处：《中国医师杂志》2022年第5期658-661,666,共5页Journal of Chinese Physician

基　　金：复旦大学附属中山医院智慧医疗专项(2020ZHZS08);复旦大学医工结合项目(XM03211181)。

摘　　要：目的通过人工分割标注腹部增强CT门脉期图像中的肝、脾和门静脉主干结构,从中提取影像组学特征,并联合临床指标构建预测乙肝肝硬化患者门脉压力的模型。方法回顾性纳入2016年1月至2020年5月于复旦大学附属中山医院就诊,行经颈静脉肝静脉压力梯度(HVPG)测定和腹部增强CT检查的乙肝肝硬化门脉高压患者。使用ITK-SNAP 3.8软件对门静脉期肝脏、脾脏和肝门静脉主干全层进行人工勾画标注。使用Python编程提取这3个部位的影像组学特征,建立HVPG预测模型。结果共纳入171例患者。年龄(51.1±10.3)岁,其中男134例(78.4%),HVPG平均值为16.9±5.7。每例患者均提取肝、脾、门静脉三个部位的影像组学特征,使用LASSO对所提取的2553个影像组学特征进行筛选,联合临床特征及影像组学特征构建HVPG的预测模型:m;VPG=31.622+0.0288×总胆汁酸-6.31(门脉主干wavelet-LHH;lcm;lusterShade)+0.253(门脉主干wavelet-LHL;lszm;argeAreaLowGrayLevelEmphasis)-20.9(脾wavelet-LLH;lcm;orrelation)-0.000127(肝original;hape;urfaceArea)+2.79(肝wavelet-LLH;lcm;lusterShade),该模型决定系数R2=0.345。结论肝、脾、门静脉影像组学特征联合临床特征可能作为乙肝肝硬化患者门脉压力无创性的评估方法。Objective In this study,the liver,spleen,and hepatic portal vein in the portal venous phase images of abdominal enhanced computed tomography(CT)are artificially segmented and annotated,and the radiomics features are extracted from them.A model for predicting portal pressure in patients with hepatitis B virus(HBV)related cirrhosis is constructed by combining radiomics features with clinical indicators.Methods A total of 171 patients who had abdominal enhancement CT examination and trans-jugular hepatic venous pressure gradient(HVPG)measurement at the same time were enrolled from January 2016 to May 2020 in the Zhongshan Hospital Affiliated to Fudan University.The liver,spleen,and hepatic portal vein in the portal venous phase images of the CT were manually labeled by using ITK-SNAP 3.8 software.The radiomics features of these three sites were extracted using Python programming,and an HVPG prediction model was established.Results A total of 171 patients was included in the study.The average age was(51.1±10.3)years,of which 134(78.4%)were males,and the average HVPG was 16.87±5.695.A total of 2553 radiomics features were extracted from three sites of the portal venous phase images of abdominal enhanced CT in each patient.The 2553 features extracted were screened using LASSO,and by combing with clinical features and radiomics features,the predictive model of HVPG was obtained:m_HVPG=31.622+0.0288T×total bile acids-6.31(portal venous wavelet-LHH_glcm_ClusterShade)=0.253(portal venous wavelet-LHL_glszm_LargeAreaLowGrayLevelEmphasis)-20.9(spleen wavelet-LLH_glcm_Correlation)-0.000127(liver original_shape_SurfaceArea)+2.79(liver wavelet-LLH_glcm_ClusterShade).The coefficient of determination R2 was 0.345.Conclusions The study suggests that radiomics features of the liver,spleen,and portal venous combined with clinical features may be used as a non-invasive method to assess the portal pressure in patients with HBV-related cirrhosis.

关 键 词：乙型肝炎 肝硬化 门静脉压 肝静脉压力梯度 预测模型 

分 类 号：R512.62[医药卫生—内科学] R575.2[医药卫生—临床医学]