基于病灶内部及边缘带的CT影像组学模型预测肝泡状棘球蚴病生物活性的研究  

Prediction value of intra-and perilesional CT radiomics on the bioactivity of hepatic alveolar echinococcosis

作　　者：侯娟 张思苗 赵梓程 徐敏 刘文亚 HOU Juan;ZHANG Simiao;ZHAO Zicheng;XU Min;LIU Wenya(Imaging Center,The First Affiliated Hospital of Xinjiang Medical University,Xinjiang 830011,China)

机构地区：[1]新疆医科大学第一附属医院影像中心,新疆乌鲁木齐830011 [2]佳能医疗系统(中国)有限公司,北京100015

出　　处：《影像诊断与介入放射学》2025年第1期16-22,共7页Diagnostic Imaging & Interventional Radiology

基　　金：国家自然科学基金(81974263);新疆医科大学研究生创新项目(CXCY2024001)。

摘　　要：目的探讨基于病灶内部及边缘带的CT影像组学模型预测肝泡状棘球蚴病(HAE)生物活性的应用价值。方法回顾性收集2012年1月—2023年12月经病理学确诊为HAE并明确生物活性的303例患者的临床资料及术前CT门脉期影像数据(有活性182例;无活性121例)。手动勾画靶病灶整体体积(GLV),并利用软件的自动扩展功能在边缘带向外扩展10、15及20 mm的范围(病灶周围总体积GPLV10~20 mm)。按8∶2的比例将入组病例随机分为训练集(n=242)和验证集(n=61)。将提取的影像组学特征经Pearson相关系数筛选以及最小绝对收缩和选择算子(LASSO)降维后,利用5种不同分类器构建影像组学模型,采用曲线下面积(AUC)、敏感度、特异度、精确度和F1分数来评估各模型预测HAE生物活性的效能,筛选出最佳分类器进一步构建不同兴趣容积的预测模型,进而确定最优组学模型。结果每个病灶区域均提取1874个影像组学特征。5种机器学习算法构建的模型均具有预测价值,其中极端随机树(ExtraTrees)算法表现最好。最优组学模型GPLV20 mm的预测效能最佳,其在训练集和验证集的AUC分别为0.876(95%CI:0.833~0.918)、0.802(95%CI:0.690~0.914)。结论联合病灶内部及边缘带的影像组学模型可更好地预测肝泡状棘球蚴病生物活性,基于病灶内部及边缘20 mm范围的影像组学特征构建的预测模型具有最好的预测效能。Objective To explore the value of intra-and perilesional radiomics based on CT in predicting bioactivity of hepatic alveolar echinococcosis(HAE).Methods CT of 303 patients with pathologically confirmed bioactive(182)and bioinactive(121)HAE from January 2012 to December 2023 was retrospectively analyzed.The gross lesion volume(GLV)was manually delineated and automatic expansion software was used to expand the gross perilesional volume(GPLV10-20 mm)outwardly by 10 mm,15 mm and 20 mm.All patients were randomly divided into training cohort(242)or test cohort(61)at a ratio of 8∶2.Feature selection was performed by intra-class correlation coefficient,Spearman correlation coefficient,and least absolute shrinkage and selection operator(LASSO).Five conventional machine learning algorithms were used to construct the radiomics prediction model.Area under receiver operating characteristic curve(AUC),sensitivity,specificity,accuracy and F1 score were used to evaluate the prediction performance of the model.Results A total of 1874 radiomics features were extracted from each volume of interest(VOI).All models constructed by the five machine learning algorithms had predictive value with best performance by the ExtraTrees algorithm.The GPLV20 mm model with an AUC of 0.876(95%CI:0.833-0.918)in the training set and 0.802(95%CI:0.690-0.914)in the validation set had the best prediction performance among the four VOIs.Conclusion The models combining intra-and perilesional radiomics can predict the bioactivity of HAE with best prediction performance by the model based on the GPLV20 mm.

关 键 词：肝泡状棘球蚴病 体层摄影术 X线计算机 机器学习 影像组学 生物活性 

分 类 号：R532.32[医药卫生—内科学] R814.42[医药卫生—临床医学]