摘要
目的观察超声影像组学结合机器学习(ML)模型鉴别肝泡型包虫病(HAE)与原发性肝癌(PHC)的价值。方法回顾性分析经手术病理证实的95例HAE(HAE组)及97例PHC(PHC组)患者,按7∶3比例将其分为训练集(n=134)及验证集(n=58);基于二维超声声像图提取影像组学特征,以逻辑回归、朴素贝叶斯、决策树、支持向量机、极限梯度提升、K邻近及梯度提升机共7种算法构建ML模型;绘制受试者工作特征曲线,计算曲线下面积(AUC),评估各模型鉴别HAE与PHC的效能。结果共基于训练集数据提取1688个影像组学特征,最终筛选出7个非零系数特征为最优特征;据以构建的7种ML模型鉴别HAE与PHC效能均良好,在训练集的AUC为0.963~0.997,在验证集为0.871~0.942。结论超声影像组学结合ML模型鉴别HAE与PHC效能良好。
Objective To observe the value of ultrasound radiomics combined with machine learning(ML)models for differentiating hepatic alveolar echinococcosis(HAE)and primary hepatic carcinoma(PHC).Methods Data of 95 HAE(HAE group)and 97 PHC patients(PHC group)confirmed by surgical pathology were retrospectively analyzed.The patients were divided into training set(n=134)and validation set(n=58)at the ratio of 7∶3.Radiomics features were extracted based on two-dimensional ultrasound,and then 7 ML models,including logistic regression,naive Bayesian,decision tree,support vector machine,extreme gradient boosting,K-nearest neighbor and gradient boosting machine were constructed.Receiver operating characteristic curves were drawn,and the area under the curves(AUC)were calculated to evaluate the efficacy of each model for differentiating HAE and PHC.Results Totally 1688 radiomics features were extracted based on training set,and finally 7 non-zero coefficient features were selected as the optimal features.All 7 ML models had good efficacy for differentiating HAE and PHC,with AUC of 0.963—0.997 in training set,of 0.871—0.942 in validation set.Conclusion Ultrasound radiomics combined with ML models had good efficacy for differentiating HAE and PHC.
作者
姑丽哥娜·乃再尔
古再努尔·阿力木
马爱琳
GULIGENA·Naizaier;GUZAINUER·Alimu;MA Ailin(Department of Ultrasound,Kashgar First People’s Hospital,Kashgar 844000,China)
出处
《中国介入影像与治疗学》
北大核心
2024年第7期423-426,共4页
Chinese Journal of Interventional Imaging and Therapy
基金
喀什地区第一人民医院“珠江学者·天山英才”合作专家工作室创新团队计划(KDYY202022)。
关键词
棘球蚴病
肝
肝肿瘤
诊断
鉴别
超声检查
影像组学
机器学习
echinococcosis,hepatic
liver neoplasms
diagnosis,differential
ultrasonography
radiomics
machine learning
作者简介
第一作者:姑丽哥娜·乃再尔(1995-),女(维吾尔族),硕士,医师。研究方向:腹部及小器官超声诊断。E-mail:3108995315@qq.com;通信作者:马爱琳,喀什地区第一人民医院超声科,844000。E-mail:kdyymal@163.com。
