采用人工智能(artificial intelligence, AI)的胸部低剂量计算机断层扫描(low-dose computed tomography, LDCT)影像特征分析肺结节良恶性的危险因素，建立肺结节良恶性预测模型。

选择2021年1月至2022年12月我院收治的肺结节患者240例，收集临床资料，采用LDCT及AI鉴别肺结节，进行多因素分析，筛选肺结节良恶性的危险因素，建立二元Logistics回归模型，比较AI、影像医师及预测模型的诊断价值。

最小CT值、直径R、毛刺征、血管穿行征、纯磨玻璃结节、部分实性结节是影响肺结节良恶性的危险因素(P<0.05)。二元Logistics回归模型为logit(P)＝－2.905＋(0.93×直径R)＋(1.572×血管穿行)＋(1.346×毛刺征)＋(1.755×纯磨玻璃结节)＋(2.25×部分实性结节)－(0.001×最小CT值)，AI、影像医师及预测模型鉴别的灵敏度分别为89.86%、81.88%、73.19%，特异度分别为32.35%、55.88%、73.53%，阳性似然比分别为1.328、1.856、2.762，阴性似然比分别为0.314、0.324、0.365，曲线下面积(area under the curve, AUC)分别为0.611、0.689、0.789。

联合肺结节形态特征及基于AI的CT定量参数回归模型对肺结节良恶性的诊断价值优于AI及影像医师，具有临床意义。

To analyze the risk factors of benign and malignant lung nodules and establish a predictive model by the low dose computed tomography (LDCT) images and artificial intelligence (AI).

The clinical data of 240 patients, who were admitted to our hospital from January 2021 to December 2022 and had definitive pathological results, were retrospectively analyzed. Univariate and multivariate analyses were performed to identify the risk factors associated with benign and malignant pulmonary nodules. A binary logistic regression model was established, and the diagnostic efficiency of AI, radiologists, and prediction models was evaluated.

Risk factors for distinguishing benign and malignant pulmonary nodules included minimum CT value, diameter R, spiculation sign, vascular penetration sign, pure ground glass nodules, and partially solid nodules(P<0.05). The regression model constructed was logit(P)=-2.905+ (0.93×diameter R)+ (1.572×vascular penetration)+ (1.346×spiculation sign)+ (1.755×pure ground glass nodules)+ (2.25×partially solid nodules)-(0.001×minimum CT value). The sensitivity of AI, radiologists, and the prediction model was 89.86%, 81.88%, and 73.19% respectively, with corresponding specificities of 32.35%, 55.88%, and 73.53%. The positive likelihood ratios was 1.328, 1.856, and 2.762, and the negative likelihood ratios are 0.314, 0.324, and 0.365. The AUC values was 0.611, 0.689, and 0.789 respectively.

The diagnostic performance of the regression model, which combines pulmonary nodule morphological features and AI-based CT quantitative parameters, in distinguishing between benign and malignant pulmonary nodules is moderate, but superior to that of AI alone or individual radiologists.