CT影像组学预测胃肠道间质瘤病理类型的诊断价值  

Diagnostic value of CT radiomics in predicting pathological types of gastrointestinal stromal tumors

作　　者：马伟丽 赵振华[1] 夏阳 黄亚男 王挺[1] MA Weili;ZHAO Zhenhua;XIA Yang;HUANG Ya’nan;WANG Ting(Department of Radiology,Shaoxing People’s Hospital,Shaoxing,Zhejiang Province 312000,China;Department of Radiology,Shaoxing Maternal and Child Health Hospital,Shaoxing,Zhejiang Province 312000,China)

机构地区：[1]绍兴市人民医院放射科,浙江绍兴312000 [2]绍兴市妇幼保健院放射科,浙江绍兴312000

出　　处：《实用放射学杂志》2023年第2期229-232,共4页Journal of Practical Radiology

基　　金：浙江省医药卫生一般研究计划课题项目(2021KY1135);浙江省医药卫生科技计划-青年人才计划项目(2019RC295)。

摘　　要：目的探讨CT影像组学模型对胃肠道间质瘤(GIST)与其他间叶源性肿瘤的鉴别诊断价值。方法回顾性选取经病理证实的胃肠道间叶源性肿瘤147例,其中GIST 105例,平滑肌瘤16例,神经鞘瘤26例。采用LASSO回归方法进行特征筛选。基于所选特征通过机器学习算法建立Logistic回归模型与随机森林(RF)模型,按照8︰2的比例分为训练集(117例)和验证集(30例)。使用受试者工作特征(ROC)曲线评价不同模型的鉴别诊断效能。结果经过特征筛选,13个影像组学特征用于模型的建立。Logistic回归模型预测的训练集曲线下面积(AUC)为0.741,验证集AUC为0.566,RF模型训练集AUC为0.935,验证集AUC为0.728,2种模型AUC比较差异有统计学意义(P<0.05)。结论基于影像组学特征构建的机器学习模型在预测GIST和非GIST病理分型上有较好的鉴别诊断价值,且RF模型鉴别价值优于Logistic回归模型。Objective To explore the value of the CT radiomics model in the differential diagnosis of gastrointestinal stromal tumors(GIST)and other mesenchymal tumors.Methods A total of 147 patients with gastrointestinal mesenchymal tumors confirmed by pathology were collected retrospectively,including 105 cases of GIST,16 cases of leiomyomas,and 26 cases of schwannomas.LASSO regression method was used for feature screening.Based on the selected features,Logistic regression model and random forest(RF)model were established by machine learning algorithm,which were divided into training set(117 cases)and validation set(30 cases)according to the ratio 8︰2.The receiver operating characteristic(ROC)curve was used to evaluate the differential diagnosis performance of different models.Results After feature screening,13 radiomics features were used to establish the model.The area under the curve(AUC)of the Logistic regression model was 0.741 in the training set and 0.566 in the validation set,and 0.935 in the RF model training set and 0.728 in the validation set.There was a significant difference in AUC between the two models(P<0.05).Conclusion The machine learning model based on radiomics features has a better differential diagnostic value in predicting GIST and non-GIST.The differential value of the RF model is better than the Logistic regression model.

关 键 词：胃肠道间质瘤 影像组学 机器学习 鉴别诊断 

分 类 号：R735[医药卫生—肿瘤] R445[医药卫生—临床医学] TP181[自动化与计算机技术—控制理论与控制工程]