分析肺结节临床与CT影像学特征，筛选出模型候选指标，构建肺结节良恶性预测模型，用于临床肺结节的筛查。

选择2014年1月至2021年12月海口市3家三甲医院收治的肺结节患者2 484例，根据术后病理结果分为恶性组710例、良性组1 774例，通过单因素和多因素Logistic回归分析患者的临床和CT影像学特征，筛选出模型候选指标，采用随机分组将2 484例患者按7︰3比例分为训练集1 739例和测试集745例，构建肺结节良恶性预测模型。

年龄、分叶征、毛刺征、胸膜牵拉征、血管集束征、晕征、空气支气管征、支气管截断征、结节周围支扩征、结节周围炎症、卫星灶、钙化、厚壁空洞、薄壁空洞两组差异有统计学意义(P<0.001)。构建回归方程P＝exp(X)/[1+exp(X)]，X＝－2.90+(0.06×年龄)+(1.95×分叶征)+(1.08×毛刺征)+(1.48×胸膜牵拉征)+(2.40×血管集束征)+(1.19×厚壁空洞)+(－1.64×薄壁空洞)+(1.14×空气支气管征)+(1.35×支气管截断征)+(－3.18×结节周围炎症)+(－0.99×卫星灶)+(1.78×晕征)+(－2.99×结节周围支扩征)+(－2.60×结节内钙化)。该模型绘制ROC曲线，AUC为0.968，95%CI：0.955~0.981，当截点值T＝1.528时，敏感度为96%，特异性为81%，阳性预测值为93%，阴性预测值为88%，准确率为92%。

年龄、分叶征、毛刺征、胸膜牵拉征、血管集束征、空气支气管征、支气管截断征、厚壁空洞和晕征是肺结节恶性的危险因素，结节周围支扩征、结节周围炎症、卫星灶、薄壁空洞、钙化是肺结节恶性的保护因素，构建的预测模型具有较高的灵敏度和特异度，可用于临床肺结节良恶性的筛查。

The clinical and CT imaging characteristics of pulmonary nodules were analyzed, the model candidates were selected, and the prediction model of benign and malignant pulmonary nodules was constructed for the clinical screening of benign and malignant pulmonary nodules.

Retrospective analysis of 2, 484 patients with pulmonary nodules admitted to the First Affiliated Hospital of Hainan Medical College, Hainan Provincial People′s Hospital and Haikou Municipal People′s Hospital from January 2014 to December 2021, According to the postoperative pathological results, it was divided into malignant group 710 case and benign group 1 774 case, The clinical and CT imaging characteristics of the patients were analyzed by univariate and multivariate Logistic regression, Screout model candidate indicators, By randomization, 2, 484 patients were divided into training set 1 739 case in a 7︰3 ratio and test set 745 case, Construct the benign and malignant prediction model of pulmonary nodules.

Age, leaf segmentation, burr, pleural pull, vascular collection, halo, air, bronchial sign, bronchial resection, perinodulular branch expansion, perinodule inflammation, satellite focus, calcification, thick-walled cavity, and thin-walled cavity (P<0.001). Build the regression equation P=exp(X)/[1+ exp(X)], X=-2.90+ (0.06 age)+ (1.95 leaf sign)+ (1.08 burr)+ (1.48 pleural pull)+ (2.40 vascular bundle sign)+ (1.19 thick wall hole)+ (-1.64 cavity)+ (1.14 air bronchial sign)+ (1.35 bronchial cutoff)+ (-3.18 surrounding nodule inflammation)+ (-0.99 satellite focus)+ (1.78 halo)+ (-2.99 peripheral branch expansion)+ (-2.60 internal nodular calcification). The model plots the ROC curve, AUC of 0.968, 95%CI of 0.955 to 0.981, when the cut-off value T=1.528, the sensitivity was 96%, specificity of 81%, positive predictive value of 93%, negative predictive value of 88% and accuracy of 92%.

Age, leaf sign, burr, pleural pull, vascular collection, air bronchial sign, bronchial cut, thick wall cavity and halo signs are malignant risk factors of lung nodules, nodules, inflammation, satellite focus, thin wall cavity, calcification is lung nodules malignant protective factors, the prediction model has high sensitivity and specificity, can be used for clinical screening of benign and malignant pulmonary nodules.