CT影像组学对亚厘米实性肺结节良恶性鉴别的应用价值
Application value of CT radiomics in differentiating malignant and benign sub-centimeter solid pulmonary nodules
刘嘉宁 1齐琳琳 1陈佳琪 1李凤兰 1崔舒蕾 1程赛楠 1王雅雯 1周振 2王建卫1
作者信息
- 1. 国家癌症中心/国家肿瘤临床医学研究中心/中国医学科学院北京协和医学院肿瘤医院影像诊断科,北京 100021
- 2. 北京深睿博联科技有限责任公司,北京 100089
- 折叠
摘要
目的 探讨CT影像组学在亚厘米实性肺结节良恶性诊断上的应用效能和潜力.方法 回顾性分析我院2020年3月—2023年 1月经增强CT检出的亚厘米(≤10mm)实性肺结节,恶性结节由手术病理证实,良性结节由手术病理或随诊证实.对病灶进行手动分割后提取影像组学特征,通过特征性相关分析和最小绝对收缩与选择算子(LASSO)算法进行特征降维,采用五折交叉验证法对模型进行验证.分别建立支持向量机、逻辑回归、线性分类支持向量机、梯度提升和随机森林组学模型并绘制受试者工作特征(ROC)曲线,采用Delong检验比较不同分类器间的诊断效能,选出效能最佳的模型与中高年资放射科医师的诊断进行对比.结果 共纳入 303例肺结节(恶性 136例),经特征提取和筛选后建立组学模型.支持向量机、逻辑回归、线性分类支持向量机、随机森林和梯度提升模型在验证集上的ROC曲线下面积分别为 0.922(95%CI:0.893,0.950)、0.910(95%CI:0.878,0.942)、0.905(95%CI:0.872,0.938)、0.899(95%CI:0.865,0.933)和 0.896(95%CI:0.862,0.930),Delong检验提示五类模型诊断效能的差异无统计学意义,其中支持向量机模型具有最高的准确率和F1分数.将支持向量机与医师诊断结果对比,其准确率高于医师(83.8%vs.55.4%,P<0.001).结论 影像组学模型对亚厘米实性肺结节良恶性诊断具有良好的诊断效能,有助于减少医师诊断的不确定性.
Abstract
Objective To investigate the application efficiency and potential of CT radiomics in differentiating malignant and benign sub-centimeter solid pulmonary nodules.Methods A retrospective study was performed on the sub-centimeter(≤10 mm)solid pulmonary nodules detected by enhanced CT in our hospital from March 2020 to January 2023.Malig-nancy was confirmed by surgical pathology,and benignity was confirmed by surgical pathology or follow-up.Lesions were manually segmented and radiomic features were extracted.The feature dimension was reduced via feature correlation analys-is and least absolute shrinkage and selection operator(LASSO).The 5-fold cross validation was used to validate the model.Support vector machine,logistic regression,linear classification support vector machine,gradient boosting,and random forest models were established for CT radiomics.Receiver operating characteristic curves were drawn.Delong test was used to compare the diagnostic performance of the five classifiers.The optimal model was selected and compared to radiologists with medium and high seniority.Results A total of 303 nodules,136 of which were malignant,were examined.Radiom-ics models were established after feature extraction and selection.On test set,the areas under the receiver operating charac-teristic curves of support vector machine,logistic regression,linear classification support vector machine,random forest,and gradient boosting models were 0.922(95%CI:0.893,0.950),0.910(95%CI:0.878,0.942),0.905(95%CI:0.872,0.938),0.899(95%CI:0.865,0.933),and 0.896(95%CI:0.862,0.930),respectively.Delong test indicated no significant differences in the performance of the five radiomics models,and the support vector machine model showed the highest accuracy and F1 score.The support vector machine model showed significantly higher diagnostic accuracy as compared to radiologists(83.8%vs.55.4%,P<0.001).Conclusion The radiomics models achieved high diagnostic efficiency and may help to re-duce the uncertainty in diagnosis of malignant and benign sub-centimeter solid nodules by radiologists.
关键词
肺肿瘤/人工智能/影像组学/鉴别诊断/电子计算机断层扫描Key words
Lung neoplasms/Artificial intelligence/Radiomics/Diagnosis and differentiation/Computed tomography引用本文复制引用
基金项目
北京市自然科学基金(7222148)
国家自然科学基金(81971616)
中国医学科学院医学与健康科技创新工程项目(2021-I2M-C&T-B-065)
出版年
2024
NETL
NSTL
万方数据