【摘要】 目的 分别应用患者的临床资料及CT影像特征、影像组学特征和两者联合的方法，构建和验证术前鉴别结节型肺黏液腺癌(PNMA)与肺炎性肌纤维母细胞瘤(PIMT)的预测模型。方法 回顾性分析2016年1月至2023年12月在本院接受手术干预的120例PNMA患者和89例PIMT患者的数据，并提取筛选影像组学特征。根据筛选出的组学特征，应用机器学习方法构建并筛选最佳影像组学模型，并在此基础上构建临床模型和临床-影像组学联合模型。采用受试者工作特征曲线(ROC)、决策曲线分析(DCA)和KS统计图，来评估各模型预测效果，Brier评分来比较各模型的整体准确性。结果 梯度提升决策树(GBDT)方法构建的联合模型效能最佳，临床模型、影像组学模型和联合模型在训练集和测试集中的ROC曲线下面积(AUC)值分别为0.983、1.000、1.000和0.885、0.901、0.923。联合模型的Brier评分为0.108。KS统计图进一步验证了该模型的卓越性，在最优预测概率阈值下，KS值为0.750。结论 利用增强CT建立的临床影像组学联合模型具有较高的预测价值，在术前鉴别PNMA和PIMT具有临床应用的潜力。更多还原

【Abstract】 Objective To create and validate a prediction model to distinguish pulmonary mucinous adenocarcinoma(PNMA) from pulmonary inflammatory myofibroblastictumor(PIMT) using clinical data, CT image features, and radiomics characteristics. Methods This retrospective study analyzed data from 120 PNMA and 89 PIMT patients who had surgery at our hospital between January 2016 and December 2023.Radiomics features were extracted, and the best radiomics model was developed using machine learning.Subsequently, clinical and combined clinical-radiomics models were constructed.Various metrics, including the receiver operating characteristic curve(ROC),decision curve analysis(DCA),and Kolmogorov-Smirnov statistical plot, were employed to assess the predictive efficacy of the three models in distinguishing between PNMA and PIMT preoperatively.The Brier score was utilized to compare the overall accuracy of each model. Results The GBDT machine learning method was utilized to construct a joint model that demonstrated superior performance, as evidenced by AUC values of 0.983,1.000,1.000 and 0.885,0.901,0.923 for the clinical model, radiomics model, and joint model in both the training and testing groups, respectively.The Brier score for the joint model was calculated to be 0.108.The KS statistical chart further validated the model’s excellence, with a KS value of 0.750 achieved under the optimal prediction probability threshold. Conclusion The clinical radiomics combined model utilizing enhanced CT demonstrates a high predictive value and holds potential for clinical application in distinguishing between PNMA and PIMT preoperatively.更多还原