摘要
目的阐述CT影像组学无创分析方法在辨别肺囊型包虫病(CE)破裂与肺脓肿中的价值。方法回顾性分析病理学证实为肺CE破裂患者40例(男25例,女15例),选取同期核查的肺脓肿患者80例(男44例,女36例),充分利用双盲法对CT图像展开观测与勾画;将病例按6∶2∶2随机分成训练集、验证集和测试集,选取4种不同的分类器(KNN、LR、RF和SVM)对筛选出的特点展开机器学习建模,通过ROC曲线下的总面积(AUC)、敏感度、特异度、准确率4个重要的指标来评价各个分类器所构建分析模型的效能。结果从CT图像中一共筛选出1409个特征,选用Lasso算法降维之后,最后选取出19个最佳特征。SVM分析模型辨别肺CE破裂患者与肺脓肿效能最佳(AUC=0.832,95%CI=0625~0.952,敏感度=0.909,特异度=0.692)。结论利用CT影像组学筛选出有价值病灶的形状特征及纹理特征可以填补肉眼观察的不足,在破裂的肺CE与肺脓肿辨别中具有重要的实际意义。
Objective To evaluate the value of CT imaging noninvasive analysis in distinguishing pulmonary cystechinococcosis(CE)rupture from lung abscess.Methods 40 patients(25 males and 15 females)with pulmonary CE rupture confirmed by pathology in our hospital from January 2010 to November 2020 were retrospectively analyzed,and 80 patients(44 males and 36 females)with pulmonary abscess were selected during the same period.Double blind method was used to observe and sketch CT images.Press the patient 6∶2∶2 randomly divided into training set,validation set and test set,select four different classifiers(KNN,LR,RF and SVM)to extract the characteristics of machine learning model,through the area under the ROC curve(AUC),the sensitivity,specific and accurate four important indexes to evaluate effectiveness of each classifier constructed analytical model.Results A total of 1409 features were screened from CT images,and 19 optimal features were selected after dimensionality reduction using Lasso algorithm.SVM analysis model had the best performance in distinguishing patients with lung CE rupture and lung abscess(AUC=0.832,95%CI=0.625-0.952,sensitivity=0.909,specificity=0.692).Conclusion Using CT image omics to screen out the shape and texture features of valuable lesions can fill in the deficiency of visual observation,and has important practical significance in the discrimination of ruptured lung CE and lung abscess.
作者
郁耀辉
刘文亚
赵圆
栗岩
刘倩
齐海成
辛娟
李杉
邢艳
YU Yaohui;LIU Wenya;ZHAO Yuan(Imaging Center,The First Affiliated Hospital of Xinjiang Medical University,State Key Laboratory of Pathogenesis,Prevention,Treatment of Central Asian High Incidence Diseases Fund,Urumchi,the Xinjiang Uygur Autonomous Region 830054,P.R.China)
出处
《临床放射学杂志》
北大核心
2022年第11期2041-2045,共5页
Journal of Clinical Radiology
基金
省部共建中亚高发病成因与预防国家重点实验室开放课题资助项目(编号:SKL-HIDCA-2020-17)。
