摘要
目的基于MRI的常规T2加权成像(T2-weighted imaging,T2WI)序列,比较采用不同机器学习算法所建立的影像组学模型在卵巢上皮性肿瘤术前三分类中的诊断效能。材料与方法回顾性分析300例(良性、交界性和恶性各100例)经病理证实为卵巢上皮性肿瘤患者的术前磁共振图像,按8∶2随机划分训练集和测试集。从轴位T2WI图像上手动勾画的三维感兴趣区域中提取图像特征,并进行特征筛选。将4种特征选择方法和7种机器学习分类器两两组合,构建28个分类模型。采用曲线下面积(AUC)和准确度对所有模型的预测性能进行评估。结果28种分类模型中表现最好的是递归特征消除法(recursive feature elimination,RFE)与K近邻(K nearest neighbor,KNN)分类器相结合的"RFE-KNN"模型,其测试集上良性组的AUC为0.94,交界性组的AUC为0.93,恶性组的AUC为0.96。结论从常规T2WI序列中提取的定量影像组学特征所建立的RFE-KNN模型在卵巢上皮性肿瘤的术前三分类中具有良好的表现。
Objective:Conventional T2WI sequences based on MRI were used to compare the diagnostic efficacy of the radiomics models established by different machine learning algorithms in preoperative tri-classification of epithelial ovarian tumors.Materials and Methods:Preoperative MR images of 300 patients(100 benign,100 borderline and 100 malignant) with pathologically confirmed ovarian epithelial tumors were retrospectively analyzed,and all the data were randomly divided into training sets and testing sets according to the ratio of 8∶2.Image features are extracted from the volume of interest(VOI) manually drawn on the axial T2WI,and screening them.Four feature selection methods and seven machine learning classifiers were pairwise combined to construct 28 classification models.AUC and accuracy were used to evaluate the prediction performance of all models.Results:The best performance among 28 classification models is the "RFE-KNN" model that combines recursive feature elimination(RFE) and K nearest neighbor(KNN) classifiers.AUC of benign,borderline and malignant group was 0.94,0.93 and 0.96.Conclusions:Quantitative radiomics features extracted from T2WI have a good performance in differentiating benign,borderline,and malignant epithelial ovarian tumors.
作者
胡艳
刘洋
郑伊能
肖智博
陈丽平
张剑
戴梦莹
李光辉
钟雨晴
马斯
吕发金
HU Yan;LIU Yang;ZHENG Yineng;XIAO Zhibo;CHEN Liping;ZHANG Jian;DAI Mengying;LI Guanghui;ZHONG Yuqing;MA Si;LÜFajin(State Key Laboratory of Ultrasound in Medicine and Engineering,College of Biomedical Engineering,Chongqing Medical University,Chongqing,400016,China;Department of Radiology,the First Affiliated Hospital of Chongqing Medical University,Chongqing 400016,China)
出处
《磁共振成像》
CAS
CSCD
北大核心
2021年第12期34-38,54,共6页
Chinese Journal of Magnetic Resonance Imaging
关键词
卵巢上皮性肿瘤
磁共振成像
影像组学
机器学习
T2加权成像
ovarian epithelial tumor
magnetic resonance imaging
radiomics
machine learning
T2-weighted imaging
