CT影像组学和机器学习预测胆囊息肉样病变性质的研究  

The Value of CT Radiomics and Machine Learning in Predicting the Nature of Gallbladder Polypoid Lesions

作　　者：尹胜男[1] 计一丁 丁宁[1] 李梦娟 迟婧 刘利[1] 张亦弛 金龙 YIN Sheng-nan;JI Yi-ding;DING Ning;LI Meng-juan;CHI Jing;LIU Li;ZHANG Yi-chi;JIN Long(Department of Medical Imaging,Suzhou Ninth Hospital Affiliated to Soochow University,Suzhou 215200,Jiangsu Province,China)

机构地区：[1]苏州大学附属苏州九院影像科,江苏苏州215200

出　　处：《中国CT和MRI杂志》2023年第12期105-108,共4页Chinese Journal of CT and MRI

基　　金：江苏省医学会伦琴影像科研专项资金项目[SYH-3201150-0018(2021013)]苏州市“科教兴卫”青年科技项目(KJXW2022077)。

摘　　要：目的探讨不同机器学习的影像组学模型预测胆固醇性胆囊息肉和腺瘤性胆囊息肉的价值。方法回顾性分析2015年9月至2022年9月我院100例经手术病理证实的胆固醇性息肉和腺瘤性息肉患者的临床及影像资料。基于术前增强CT提取的影像组学特征,训练集采用T检验和最小绝对收缩和选择算子交叉验证法进行特征筛选。然后用3种机器学习方法(人工神经网络、逻辑回归和支持向量机)构建预测模型,利用ROC曲线下面积、准确率以及F1度量值评估3种模型的预测能力,并通过验证集进行验证。结果人工神经网络算法在基于10个影像组学特征的息肉病变性质鉴别模型中预测效果最好,具有最高的曲线下面积(训练集为1;验证集为0.81)、准确率(训练集为1;验证集为0.83)及F1度量值(训练集为1;验证集为0.76)。10个影像组学特征中,基本特征1个、小波特征9个。结论基于增强CT的机器学习影像组学模型有助于预测胆固醇性胆囊息肉和腺瘤性胆囊息肉的病变性质,为两者的定性诊断及治疗方式的选择提供依据。Objective To explore the value of different radiomics models based on machine learning in predicting cholesterol and adenomatous gallbladder polyps that were inaccurately diagnosed before surgery.Methods The clinical and imaging data of 100 patients with cholesterol polyps and adenomatous polyps confirmed by surgery and pathology in Suzhou Ninth People's Hospital from September 2015 to September 2022 were analyzed retrospectively.Based on the preoperative dynamic contrast-enhanced CT radiomics features,T test and least absolute shrinkage and selection operator cross validation were applied to select features.Then,three machine learning methods(artificial neural network,logistic regression and support vector machine)are used to build prediction models,and the area under the ROC curve,accuracy and F1 measure are used to evaluate the prediction ability of the three models,which are verified by the validation group.Results The prediction effect of artificial neural network algorithm was the best in the model of polyp pathological properties identification based on 10 radiomics features,with the highest area under the curve(1 in the training group;0.81 in the validation group),accuracy(1 in the training group;0.83 in the validation group),and F1 measure(1 in the training group;0.76 in the validation group).Among the 10 radiomics features,there are 1 first-order feature and 9 wavelet features.Conclusion The machine learning radiomics model based on enhanced CT is helpful to predict the characteristics of cholesterol gallbladder polyps and adenomatous gallbladder polyps,and provides a more reliable basis for their preoperative diagnosis and treatment.

关 键 词：胆囊息肉样病变 增强CT成像 影像组学 机器学习 

分 类 号：R575.6[医药卫生—消化系统]