基于能谱CT增强图像的支持向量机模型在鉴别肺实性结节良恶性的价值

The Value of SVM Model Based on Spectrum CT Enhanced Images in Distinguishing Benign and Malignant Pulmonary Solid Nodules

目的探讨基于能谱CT增强图像的支持向量机(SVM)模型在鉴别肺内实性结节良恶性的价值。方法回顾性分析2021年3月至2023年6月滨州医学院附属医院经手术或穿刺病理证实的肺内实性结节234例,其中恶性结节140例,良性结节94例。术前均行能谱CT增强扫描,应用3D-slicer软件在动脉期70 keV图像上手动勾画感兴趣区(ROI)并提取影像组学特征,应用最小绝对收缩和选择算子(LASSO)算法及互信息法降维并提取最优影像组学特征,按照7∶3的比例分为训练集及验证集,采用SVM法构建预测模型,使用受试者工作特征(ROC)曲线评估模型的诊断效能。结果共提取出1168个影像组学特征,经LASSO降维后筛选出19个特征,然后采用互信息法提取出对模型影响最大的15个特征,应用SVM观测该15个特征数据。经回归分析,构建肺实性结节的预测模型,该模型ROC曲线的曲线下面积(AUC)在训练集及验证集上分别为0.932、0.810。结论应用基于能谱CT增强图像的SVM方法在预测肺实性结节良恶性中具有较高的价值。

Objective Exploring the value of SVM model based on Spectrum CT enhanced images in distinguishing benign and malignant pulmonary solid nodules.Methods Retrospective analysis of 234 cases of solid pulmonary nodules confirmed by surgery or puncture pathology at the Affiliated Hospital of Binzhou Medical College from March 2021 to June 2023,including 140 malignant nodules and 94 benign nodules.Prior to surgery,Spectrum CT enhanced scans were performed on all images.The ROI was manually delineated and radiomics features were extracted on the arterial phase 70 kev images using 3D slicer software.LASSO algorithm and Mutual information method were used to reduce dimensionality and extract the optimal radiomics features.The model was divided into training and validation sets in a 7∶3 ratio.The SVM method was used to construct the prediction model,and the ROC curve was used to evaluate the diagnostic performance of the model.Results A total of 1168 radiomics features were extracted,and after LASSO dimensionality reduction,19 features were selected.Then,the mutual information method was used to extract the 15 features that had the greatest impact on the model.SVM was used to observe these 15 feature data.After regression analysis,a predictive model for pulmonary solid nodules was constructed,and the AUCof the ROC curve of the model was 0.932 and 0.810 in the training and validation groups.Conclusion The application of SVM method based on Spectrum CT enhanced images has high value in predicting the benign and malignant pulmonary solid nodules.