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

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(≤10 mm)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。

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.Malignancy 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 analysis 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.Radiomics models were established after feature extraction and selection.On test set,the areas under the receiver operating characteristic 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 reduce the uncertainty in diagnosis of malignant and benign sub-centimeter solid nodules by radiologists.