影像组学分析和机器学习在肺结节良恶性分类中的应用

目的:探讨CT纹理特征对良恶性肺结节的鉴别价值及在独立数据集上的泛化能力.方法:回顾性分析LIDC-IDRI和LUNGx数据库中共1428个肺结节(直径3～30 mm)的CT图像,其中良性1221个、恶性207个.将LIDC-IDRI数据库的1372个结节(良性1190个,恶性182个)作为训练集,LUNGx数据库的5...

Radiomics analysis and machine learning for classification of benign and malignant pulmonary nodules

Objective:The aim of this study was to investigate the value of CT radiomics texture features in classification of benign and malignant pulmonary nodules and its generalizability in independent datasets.Methods:This retrospective study contained 1428 pulmonary nodules(1221 benign and 207 malignant)with diameter of 3~30mm in two public datasets named LIDC-IDRI and LUNGx.The training cohort was composed of 1372 nodules(1190 benign and 182 malignant)from the LIDC-IDRI dataset and the validation cohort was composed of 56 nodules(31 benign and 25 malignant)from the LUNGx dataset.A total of 946 radiomics features were extracted from each nodule using the software package Pyradiomics.The radiomics features with significant differences between benign and malignant nodules were first identified,and then LASSO algorithm or triad method(Fisher+POE+ACC+MI,FPM)were applied for further feature selection.Finally,the classification model for pulmonary nodules was constructed using support vector machine.The performance of the optimal model was evaluated directly in validation and training cohort with cross validation procedure.Results:Using trai-ning cohort with cross validation,the AUC of the optimal model was 0.892,and the accuracy,sensitivity,specificity,positive predictive value(PPV)and negative predictive value(NPV)was 0.859,0.788,0.876,0.492 and 0.964,respectively.17 features were retained after feature selection.In validation cohort,the AUC was 0.765,and the accuracy,sensitivity,specificity,PPV and NPV were 0.745,0.800,0.700,0.689 and 0.808,respectively.Conclusion:CT radiomics texture features show good performance and generalizability in classification of malignant and benign pulmonary nodules,and that a promising approach in computer-aided diagnosis of lung cancer in clinical practice.