探讨基于CT平扫影像组学和形态学特征的列线图在术前区分肺腺癌和良性结节的应用价值。

回顾性收集2018年1月至2024年12月我院收治的248例肺结节患者，按7︰3随机分为训练集173例和验证集75例。基于肺CT平扫图像提取影像组学特征，计算肺结节的组学评分(radiomics score, Rads)；评估肺结节的形态学特征。通过多因素Logistic回归筛选危险因素，构建列线图模型。采用受试者工作特征曲线下面积(area under the curve, AUC)及可信区间(confidence interval, CI)、校准曲线和决策曲线判读模型的效能。

248例中肺腺癌142例，良性结节106例，筛选出11个关键影像组学特征及2个形态学特征(浅分叶、血管聚集)。多因素Logistic回归显示，浅分叶(比值比[odds ratio, OR]＝3.342，P＝0.002)、血管聚集(OR＝2.229，P＝0.040)和Rads(OR＝3.347，P<0.001)是术前区分肺腺癌与良性结节的预测因素。基于上述变量构建的列线图模型在训练集中AUC为0.831(95%CI: 0.767~0.884)，敏感度为67.0%，特异度为84.3%；验证集中AUC为0.825(95%CI: 0.721~0.903)，敏感度为74.4%，特异度为86.1%。校准曲线显示模型校准能力良好，训练集Hosmer-Lemeshow检验P＝0.9434，验证集P＝0.6345，决策曲线提示模型具有良好的临床净获益。

基于CT影像组学评分、浅分叶和血管聚集构建的列线图模型在术前区分肺腺癌与良性结节中具有稳定且良好的诊断性能，可为肺结节患者的个体化术前决策提供无创辅助工具。

To investigate the value of a nomogram based on non-contrast CT radiomics and morphological features in preoperatively differentiating lung adenocarcinoma from benign pulmonary nodules.

A total of 248 patients with pulmonary nodules admitted to our hospital from January 2018 to December 2024 were retrospectively enrolled and randomly divided into a training set 173 cases and a validation set 75 cases at a 7︰3 ratio. Radiomics features were extracted from non-contrast CT images. The radiomics score (Rads) was calculated from selected features. Morphological features of the pulmonary nodules were also evaluated. Multivariate logistic regression was applied to identify independent risk factors, and a nomogram model was constructed. Model performance was assessed using area under the curve of receiver operating characteristic curves (AUC), confidence interval (CI), calibration curves, and decision curve analysis.

Among the 248 cases, there were 142 lung adenocarcinomas and 106 benign nodules. Eleven key radiomics features and two morphological features (shallow lobulation and vascular convergence) were selected. Multivariate logistic regression revealed that shallow lobulation (OR=3.342, P=0.002), vascular convergence (OR=2.229, P=0.040), and Rads (OR=3.347, P<0.001) were independent risk factors for differentiating lung adenocarcinoma from benign nodules. The nomogram constructed based on these variables achieved an AUC of 0.831 (95%CI: 0.767~0.884) with a sensitivity of 67.0% and specificity of 84.3% in the training set, and an AUC of 0.825 (95%CI: 0.721~0.903) with a sensitivity of 74.4% and specificity of 86.1% in the validation set. Calibration curves demonstrated good model calibration (Hosmer-Lemeshow test: P=0.9434 in the training set, P=0.6345 in the validation set), and decision curve analysis indicated satisfactory clinical net benefit.

The nomogram model incorporating the CT radiomics score, shallow lobulation, and vascular convergence demonstrates stable and robust diagnostic performance in preoperatively differentiating lung adenocarcinoma from benign nodules. It may serve as a non-invasive auxiliary tool to support individualized preoperative decision-making for patients with pulmonary nodules.