基于CT和18F-FDG PET/CT的肺癌风险预测模型对肺结节恶性风险的验证研究

目的 比较基于CT的Brock模型、VA模型和基于18F-FDG PET/CT的Herder模型预测肺结节恶性风险的效能，验证模型的预测准确率。 方法 回顾性分析2009年7月至2016年7月行CT检查并经病理确诊或随访确诊的120例肺结节患者，其中可能为恶性的59例患者接受了18F-FDG PET/CT检查。绘制3种模型的受试者工作特征（ROC）曲线并计算曲线下面积（AUC）。测算肺结节（直径4~30 mm）的恶性风险。根据每个模型基于排除标准的队列和所有患者总队列的AUC，验证模型的准确性。采用MedCalc软件进行相关性分析，DeLong方法进行两条ROC的比较。 结果 120例肺结节患者中，49例患者（40.8％）患有恶性结节（31.6％原发性肺癌、8.2％转移性疾病）。在受排除标准限制的队列中，Brock和VA模型的AUC分别为 0.887和0.758，两者间的差异有统计学意义（Z=6.483，P=0.006）。 在接受18F-FDG PET/CT检查的患者中，Herder模型的AUC为0.937。当对队列中的所有患者（即包括原模型纳入标准之外的患者）测试模型时，每个模型的AUC值均有所降低，但两类队列比较差异无统计学意义（Z=21.357，P=0.121）。 对于≤10 mm的结节，Brock和VA 模型的AUC值分别为0.846和0.536，Brock模型明显优于VA 模型（Z=8.768，P=0.0026）。 结论 Brock模型可预测CT扫描中检测到的肺结节恶变的可能性，在接受18F-FDG PET/CT进行肺结节评估的患者中，Herder模型的预测效能最高。

Verification of malignant risk of pulmonary nodules based on CT and 18F-FDG PET/CT prediction model

Objective To compare the efficacy of the CT and 18F-FDG PET/CT models in predicting the malignant risk of pulmonary nodules and to verify the predictive accuracy of model. Methods A retrospective analysis of 120 patients with pulmonary nodules confirmed by pathological diagnosis or follow-up were conducted in this study. Among these patients, 59 patients with suspected malignancy received 18F-FDG PET/CT. The corresponding receiver operating characteristic curve for each model was plotted, and the area under the curve(AUC) was calculated. The malignant risk of patients with pulmonary nodules(4–30 mm in diameter) was measured. Model accuracy was verified based on the exclusion criteria for each model and the total cohort of all patients. MedCalc software was used for correlation analysis, and DeLong method was used for two-way comparison. Results All 120 patients with pulmonary nodules were examined. Among them, 49(40.8%) had malignant nodules(31.6% primary lung cancer and 8.2% metastatic disease). The AUC of the Brock and VA models were 0.887 and 0.758, respectively, the difference was statistically significant(Z=6.483, P=0.006). In patients receiving 18F-FDG PET/CT, the AUC of the Herder model was 0.937, which was significantly more accurate than those of the other two models. When testing the model for all patients in the cohort(i.e., patients including the original model’s inclusion criteria), the AUC value decreased but was not significant. For the Herder model, the AUC was 0.923, and the two types of cohorts were not significant(Z=21.357, P=0.121). For subcentimeter nodules, the AUC values for the Brock and VA models were 0.846 and 0.536, respectively, and the Brock model was significantly better than the VA model(Z=8.768, P=0.0026). Conclusion The Brock model showed good accuracy and was used to predict the likelihood of malignancy in nodules detected by CT scan. The Herder model was the most accurate for patients who underwent 18F-FDG PET/CT for nodule evaluation.