单发肺部小结节良恶性诊断的预测模型

A predictive model for the diagnosis of benign and malignant single lung nodule

目的:通过构建单发肺部小结节良恶性质诊断的预测模型,以提高临床早期鉴别肺部恶性结节的准确性。方法:回顾性总结2017年4月至2020年4月入我院判断为单发肺部小结节患者112例(结节直径8 mm~3 cm),经64层螺旋CT增强扫描和重建处理,良恶性经CT引导穿刺活检或者手术病理证实,其中恶性结节40例(恶性组),良性结节72例(良性组)。采用单因素分析比较两组患者性别、年龄、结节位置、结节直径、结节内部和外部纹理特征,其中内部特征包括分叶、空洞、钙化、血供,外部纹理特征包括毛刺、不规则、支气管征和血管束。多因素Logistic回归分析筛选影响性质判断的危险因素,绘制列线图预测模型、受试者工作特征曲线(ROC),计算模型预测的准确性,Hosmer-Lemeshow检验计算预测效能。结果:恶性组患者的年龄比良性组增加,结节内部和外部纹理特征数量增多(P<0.05),但两组患者性别、结节位置、结节直径比较差异不明显(P>0.05)。Logistic回归分析筛选发现,患者年龄、结节直径、结节内部和外部纹理特征数量是影响性质判断的独立危险因素(P<0.05)。列线图预测模型显示,患者年龄和结节直径越大,结节内部和外部纹理特征数量越多,提示恶性结节的可能性就越高。ROC分析显示,该模型预测的准确性为0.896;Hosmer-Lemeshow检验计算预测效能为91.07%,拟合效果良好。结论:经64层螺旋CT增强扫描和重建处理单发肺部小结节,构建恶性结节性质判断的预测模型主要指标包括患者年龄、结节直径、结节内部和外部纹理特征数量,有较好的准确性和应用价值。

Objective:To improve the accuracy of early clinical diagnosis of malignant single lung nodules by constructing a predictive model for the diagnosis of benign and malignant single lung nodule.Methods:A retrospective summary of 112 patients who were admitted to our hospital from April 2017 to April 2020 and judged to be a single pulmonary nodule(nodule diameter 8 mm-3 cm),after 64-slice spiral CT enhanced scanning and reconstruction processing,confirmed by CT-guided needle biopsy or surgical pathology,confirmed including 40 cases of malignant nodules(malignant group)and 72 cases of benign nodules(benign group).Single factor analysis was used to compare the gender,age,nodule location,nodule diameter,internal and external texture features of the nodule,the internal features including lobes,cavities,calcification,blood supply,and external texture features including burrs,irregularities,bronchial signs and vascular bundles.Multi-factor Logistic regression analysis screened out the risk factors that affected the nature of judgment.manomogram prediction model showed that the greater the patient age and nodule diameter,the greater the number of nodule internal and external markings,the higher the likelihood of malignant nodules.The ROC analysis showed that the prediction accuracy of this model was 0.896,and the Hosmer-Lemeshow test was 91.07%,with good fit.Results:The age of the patients in malignant group was higher than that in benign group,and the number of internal and external texture features of nodules increased(P<0.05),but there was no significant difference in gender,nodule location and nodular diameter between the two groups(P>0.05).Logistic regression analysis showed that patient’s age,nodule diameter and the number of internal and external texture features of nodules were independent risk factors that affecting the judgment of nature(P<0.05).Nomogram prediction model shows that the larger the patient’s age and nodule diameter,the more the number of internal and external texture features of the nodule,suggesting the higher the probability of malignant nodule.ROC analysis shows that the prediction accuracy of the model is 0.896;the prediction efficiency of Hosmer-Lemeshow test is 91.07%,and the fitting effect is good.Conclusion:After 64-slice spiral CT enhanced scanning and reconstruction,a single small lung nodule has been processed to construct a predictive model for judging the nature of malignant nodules.The main indicators include patient age,nodule diameter,and the number of internal and external texture features of the nodule with good accuracy and application value.