摘要
目的建立不同机器学习算法的增强后T1加权图像影像组学模型,并对比不同模型鉴别肺癌与非肺癌脑转移瘤的诊断效能。材料与方法将728例肺癌脑转移瘤与126例非肺癌脑转移瘤患者按照7∶3比例随机分为训练集599例与验证集255例,所有患者增强T1加权图像导入ITK-SNAP软件,手动勾画感兴趣区(region of interest,ROI)。基于ROI进行影像组学特征提取并使用最小绝对收缩选择算子进行特征筛选。基于显著特征,分别建立支持向量机(support vector machines,SVM)、随机森林(random forest,RF)和逻辑回归(logistics regression,LR)模型,采用受试者工作特征(receiver operating characteristic,ROC)曲线评估模型对肺癌脑转移瘤及非肺癌脑转移瘤的鉴别诊断效能。结果经过特征筛选后最终保留5个显著特征,诊断效能最好的SVM影像组学模型在训练集中的ROC曲线下面积(area under the curve,AUC)为0.796,准确度为85.3%,敏感度为87.8%,特异度为70.8%,验证集中的AUC为0.789,准确度为90.2%,敏感度为95.4%,特异度为59.5%。结论基于增强MR影像组学模型可用于预测原发灶不明脑转移瘤的肺癌与非肺癌原发灶肿瘤类型,SVM模型诊断价值高于RF及LR模型。
Objective:To establish radiomics of different machine learning algorithms models based on enhanced T1-weighted images,and to explore the value of the models for distinguishing lung cancer brain metastases and non-Lung cancer brain metastases.Materials and Methods:Totally 728 patients with lung cancer brain metastases and 126 patients with non-Lung cancer brain metastases were randomly divided into training set(n=599)and testing set(n=255)according to the ratio of 7∶3.Enhanced MRI data were imported into ITK-SNAP software,and the tumor's region of interest(ROI)in the enhanced T1WI was manually delineated to ROI.Radiomics feature extraction and screening using the least absolute shrinkage selection operator based on ROI.Support vector machine(SVM)model,random forest model and logistic regression model based on salient features were established respectively.Receiver operating characteristic(ROC)curve was used to assess the diagnostic efficiency of the models for distinguishing lung cancer brain metastases and non-Lung cancer brain metastases.Results:After feature screening,5 salient features were finally retained.The most effective radiomics model was the SVM model.In the training set.The area under the curve(AUC)of SVM was 0.796,accuracy value of 85.3%,sensitivity value of 87.8%,specificity value of 70.8%.In the testing set,AUC value of 0.789,accuracy of 90.2%,sensitivity of 95.4%,specificity of 59.5%.Conclusions:Radiomics models based on enhanced MRI can be used for effectively predicting the lung cancer and non-lung cancer primary focus of brain metastatic cancer with unknown primary tumor.SVM model has higher diagnostic value than random forest model and logistic regression models.
作者
隋莲玉
任嘉梁
王佳宁
殷小平
SUI Lianyu;REN Jialiang;WANG Jianing;YIN Xiaoping(Department of Radiology,Affiliated Hospital of Hebei University,Baoding 071000,China;General Electric Pharmaceutical(Shanghai)Co.,Ltd.,Shanghai 200203,China)
出处
《磁共振成像》
CAS
CSCD
北大核心
2022年第12期74-80,共7页
Chinese Journal of Magnetic Resonance Imaging
基金
河北大学研究生创新资助项目(编号:HBU2022ss024)。
关键词
脑转移瘤
肺癌
影像组学
对比增强T1加权图像
磁共振成像
brain metastases
lung cancer
radiomics
contrast-enhanced T1-weighted images
magnetic resonance imaging
