肺结节超高分辨率CT靶扫描影像组学联合机器学习在原位腺癌与微浸润腺癌鉴别中的价值  

The value of ultra-high resolution CT target scan imaging omics of pulmonary nodules combined with machine learning in distinguishing in situ adenocarcinoma from microinvasive adenocarcinoma

作　　者：胡栩晟 郭艺帆 李鲁[2] 陈晓珺 Hu Xusheng

机构地区：[1]浙江省中医院,310003 [2]浙江省金华市中心医院,321099

出　　处：《浙江临床医学》2023年第9期1294-1296,共3页Zhejiang Clinical Medical Journal

基　　金：金华市科技重点项目(2021-3-041)。

摘　　要：目的基于肺结节超高分辨率CT(UHRCT)靶扫描影像组学特征,分别采用Logistic回归(LR)和支持向量机(SVM)构建机器学习模型,以鉴别磨玻璃结节(GGN)中的原位腺癌(AIS)和微浸润腺癌(MIA)。方法回顾性分析手术病理证实肺腺癌的198例患者(AIS 56例;MIA 142例),按分层抽样将患者随机分为训练组(n=138)和验证组(n=60)。手动分割GGN,从中提取影像组学特征。采用最小冗余最大相关性算法和套索算法对影像组学特征进行降维,分别使用LR和SVM构建预测模型。采用受试者操作特征(ROC)曲线评价模型的预测性能。结果在训练组中,LR和SVM曲线下面积(AUC)分别为0.787(95%CI:0.712~0.863)和0.896(95%CI:0.842~0.951)。在验证组中,LR和SVM的AUC分别为0.824(95%CI:0.713~0.936)和0.839(95%CI:0.734~0.945)。结论基于肺结节UHRCT靶扫描影像组学结合机器学习能较好鉴别AIS与MIA,为患者GGN个性化分析提供潜在方法。Objective To discriminate between adenocarcinoma in situ(AIS)and minimally invasive adenocarcinoma(MIA)within groundglass nodules(GGN),machine learning models were constructed using logistic regression(LR)and support vector machine(SVM)based on radiomics features extracted from ultra-high-resolution CT(UHRCT)target scan images.Methods 198 patients with surgically and pathologically confirmed lung adenocarcinoma were retrospectively included(AIS 56 cases;MIA 142 cases),patients were randomly divided into a training group(n=138)and a validation group(n=60)according to stratified sampling.Manually segmented the GGN and extracted the radiomics features from it.The minimum redundancy and maximum relevance algorithm and the Lasso algorithm were used to reduce the dimensionality of the radiomics features,and the prediction model was constructed using LR and SVM respectively.The predictive performance of the model was evaluated by receiver operating characteristic(ROC)curve.Results In the training group,the area under the curve(AUC)of LR and SVM were 0.787(95%CI:0.712~0.863)and 0.896(95%CI:0.842~0.951),respectively.In the validation group,the AUC of LR and SVM were 0.824(95%CI:0.713~0.936)and 0.839(95%CI:0.734~0.945),respectively.Conclusion UHRCT target scan of pulmonary nodules-based radiomics combined with machine learning can identify AIS and MIA well,providing a potential method for GGN personalized analysis in patient.

关 键 词：肺结节 影像组学 超高分辨率CT 靶扫描 

分 类 号：R734.2[医药卫生—肿瘤] R730.44[医药卫生—临床医学]