CT征象用于孤立性肺结节良恶性鉴别诊断的Logistic回归分析及预测模型构建

摘要

Objective:To establish the Logistic regression prediction model on CT characteristics in solitary pulmonary nodule differential diagnosis, and to evaluate the diagnostic value of Logistic model in differential diagnosis between benign and malignant pulmonary nodule. Methods:Retrospective selected 188 solitary pulmonary nodule people, among them 62 cases was benign nodules and 126 cases was malignant nodules. The CT characteristics of benign and malignant solitary pulmonary nodule cases were compared, a Logistic model was obtained on the base of CT characteristics, ROC curve was constructed to assess the performance of the Logistic model. Results:Multi factor Logistic regression analysis showed that leaf sign (OR=7.085), pleural indentation sign (OR=13.224), spicule sign (OR=15.062) were the main features of the differential diagnosis of benign and malignant SPN. The predictive accuracy of Logistic regression model was 93.1% (175/188), sensitivity 91.3% (115/126), specificity 96.8% (60/62), positive predictive value 98.3% (115/1117), positive predictive value (SPN), and negative predictive value of 84.5% (115/126), negative predictive value of (60/71). The area under the ROC curve was (0.943±0.017), P<0.001, 95%CI:0.911-0.974. Conclusion:Logistic prediction model based on CT features for the differential diagnosis of benign and malignant SPN has a high value.%目的:建立基于CT影像特征预测孤立性肺结节(solitary pulmonary nodule,SPN)良恶性的Logistic回归模型,并探讨该预测模型在鉴别诊断SPN良恶性中的应用价值.方法:回顾性收集188例SPN患者,其中良性结节62例、恶性结节126例,比较两组CT征象指标,以病理诊断作为金标准,建立Logistic回归预测模型,计算预测模型准确率、灵敏度、特异度等指标,绘制ROC曲线并计算曲线下面积.结果:单因素及多因素Logistic回归分析结果显示:分叶征(OR=7.085)、胸膜凹陷征(OR=13.224)、毛刺征(OR=15.062)是SPN良、恶性鉴别诊断的主要CT影像特征指标.Logistic回归模型对SPN良恶性的预测正确率为93.1%(175/188)、灵敏度91.3%(115/126)、特异度96.8%(60/62)、阳性预测价值98.3%(115/1117)、阴性预测价值84.5%(60/71).ROC曲线下面积为(0.943±0.017),P<0.001,95%CI:0.911-0.974.结论:基于CT征象的Logistic预测模型对于鉴别SPN良恶性具有较高的价值.