基于CT平扫的影像组学模型鉴别肺炎性肌纤维母细胞瘤和周围型肺癌的研究

目的探讨基于CT平扫的影像组学在鉴别肺炎性肌纤维母细胞瘤(IMT)和周围型肺癌(PLC)的价值。方法回顾性收集经我院及外院手术后病理证实的72例肺IMT及79例PLC的资料。使用A.K(Artificial Intelligence Kit)软件从CT平扫图像中提取高通量数据,对其进行特征筛选及降维,去除了重复性差、冗余度高的特征。将数据按照7∶3∶3比例分为训练集、验证集及测试集,其中外院数据作为测试集。采用逻辑回归、支持向量机、随机森林机器算法对所提取较优特征建立分类预测模型,绘制受试者操作曲线(ROC),计算受试者操作曲线下面积(AUC),评估模型的诊断预测效能,采用Delong检验比较模型间的效能差异。采用测试集对三种机器学习模型进行评估,并绘制ROC曲线。结果共提取396纹理特征,通过特征选择及降维获得12个较优的纹理特征,三种分类模型符合率分别为77.1%、62.9%、82.9%,特异度分别为61.1%、55.6%、83.3%,敏感度分别为94.1%、70.6%、72.4%,AUC值分别为0.791、0.748、0.859,Delong检验比较模型间ROC-AUC值无统计学意义,随机森林符合率(82.9%)更高。测试集三种模型AUC值分别为0.807、0.739、0.781。结论基于CT平扫建立的影像组学特征模型有助于区分肺IMT和PLC,为临床准确诊断和个体化治疗提供客观依据。

Differential diagnosis of pulmonary inflammatory myofibroblastic tumor and peripheral lung cancer based on plain CT radiomics model

Objective To explore the value of plain CT radiomics in the differential diagnosis of pulmonary inflammatory myofibroblastic tumor(IMT)and peripheral lung cancer(PLC).Methods The plain chest CT of histologically confirmed pulmonary IMT(72)and PLC(79)was retrospectively analyzed.Using Artificial Intelligence Kit software to extract high-throughput data from CT data,feature screening and dimensionality reduction were carried out and the features of poor repeatability and high redundancy were removed.According to the proportion of 7:3:3,the data were divided into training,verification,and test sets.Logical regression,support vector machine,and random forest machine algorithms were used to establish classification prediction models for the extracted better features.The test set was used to evaluate the three machine learning models.Receiver operating characteristic(ROC)curve analysis was performed to evaluate the diagnostic efficiency of the model.Delong test was used to compare the performance differences between the models.Results A total of 396 texture features were extracted and 12 better texture features were obtained by feature selection and dimensionality reduction.The areas under the ROC curves(AUCs)of the three classification prediction models(0.791,0.748,0.859)were not statistically significant.The diagnostic accuracy of random forest machine algorithm(82.9%)with 83.3%specificity and 72.4%sensitivity was significantly higher than that of the other 2 classification prediction models(77.1%,62.9%)with 61.1%and 55.6%specificities,94.1%and 70.6%sensitivities.The AUC values of the three models of the test set were 0.807,0.739 and 0.781,respectively.Conclusion The imaging feature model based on plain CT is helpful to distinguish pulmonary IMT from PLC and provides objective basis for accurate clinical diagnosis and individualized treatment.