基于多期增强CT影像组学模型预测肝细胞癌切除术后复发的价值  

Multiphase Enhanced CT-Based Radiomics for Predicting Recurrence in Patient with Hepatocellular Carcinoma After Tumor Resection

作　　者：杨春燕 魏小琴 杨琼[4] 李杨[1] 杜勇[1] YANG Chunyan;WEI Xiaoqin;YANG Qiong;LI Yang;DU Yong(Department of Radiology,the Affiliated Hospital of North Sichuan Medical College,Nanchong 637000,China;不详)

机构地区：[1]川北医学院附属医院放射科,四川南充637000 [2]武胜县人民医院医学影像科,四川广安638400 [3]川北医学院医学影像学院,四川南充637000 [4]重庆大学附属人民医院,重庆市人民医院,重庆400010

出　　处：《中国医学影像学杂志》2025年第3期245-251,共7页Chinese Journal of Medical Imaging

基　　金：南充市市校合作科研专项(NSMC20170301)。

摘　　要：目的探讨影像组学-临床联合模型预测肝细胞癌肿瘤切除术后复发的价值。资料与方法回顾性收集2015年1月—2022年6月于川北医学院附属医院行根治性手术切除的肝细胞癌患者311例,按7∶3随机分为训练组217例和验证组94例。于动脉期、门静脉期CT图像中,对肿瘤及瘤周5 mm勾画感兴趣区,提取并筛选影像组学特征,建立影像组学模型。采用单因素和多因素Cox分析筛选临床独立危险因素,建立临床模型。将临床独立危险因素与影像组学特征联合,进而建立影像组学-临床联合模型。利用受试者工作特征曲线下面积、特异度、敏感度、净重分类改善指数以及综合判别改善指数评估模型的区分度,通过校准曲线评估模型的校准程度,运用临床决策曲线分析评估模型的临床效用,利用验证组数据对模型进行验证。结果所有模型中,Rad-A5V5-临床模型对肝细胞癌切除术后复发的预测性能最佳,验证组的曲线下面积、特异度、敏感度分别为0.743(95%CI 0.640~0.846)、0.647、0.717;净重分类改善指数和综合判别改善指数显示,与临床模型和影像组学模型比较,RadA5V5-临床模型预测能力改善效果最佳;校准曲线显示,Rad-A5V5-临床模型预测值与真实值符合度最好;决策曲线结果显示,Rad-A5V5-临床模型在特定的阈值范围内能够获得最大的临床净收益。结论瘤周影像组学特征与临床独立危险因素结合,可显著提高对肝细胞癌肿瘤切除术后复发的预测效果,为识别复发高危患者和制订个体化治疗方案提供重要参考依据。Purpose To develop and validate a radiomics-clinical model that could accurately predict the recurrence of hepatocellular carcinoma(HCC)after undergoing tumor resection.Materials and Methods A total of 311 HCC patients underwent tumor resection in the Affiliated Hospital of North Sichuan Medical College from January 2015 to June 2022 were retrospectively collected,and they were randomly divided into a training cohort(n=217)and a validation cohort(n=94)in the ratio of 7∶3.Tumor and peritumoral 5 mm regions of interest were outlined on arterial and portal venous phase images and radiomics features were extracted to establish an arterio-portal radiomics model.Independent clinical risk factors associated with postoperative recurrence were explored by univariate and multivariate Cox analysis,then clinical model was established.A combined radiomics-clinical model was established by combining clinical independent risk factors and radiomics features.The area under the curve,specificity,sensitivity,net reclassification improvement and integrated discrimination improvement were used to assess the discrimination of all models.The calibration curve assessed the calibration of the model and decision curve analysis assessed the clinical utility of the model.Validation was performed by validation cohort data.Results The Rad-A5V5-clinical model had the best predictive performance for postoperative recurrence of HCC,the area under the curve,specificity and sensitivity of the validation cohort were 0.743(95%CI 0.640-0.846),0.647 and 0.717,respectively.The results of net reclassification improvement and integrated discrimination improvement showed that compared with clinical model and radiomics model,the prediction ability of Rad-A5V5-clinical model was improved the best one.The calibration curves showed that the predicted values of the Rad-A5V5-clinical model conformed the most favorably to the true values.The results of the decision curve analysis curve analyses showed that,among all models,the Rad-A5V5-clinical model would obt

关 键 词：癌 肝细胞 复发 体层摄影术 X线计算机 肿瘤切除术 影像组学 危险因素 

分 类 号：R445.3[医药卫生—影像医学与核医学] R735.7[医药卫生—诊断学]