18F-FDG PET/CT孤立性肺结节恶性风险预测模型的建立及效能评价

A model of malignant risk prediction for solitary pulmonary nodules on 18F-FDG PET/CT: building and estimating

目的构建一种含肿瘤代谢因素的孤立性肺结节(SPN)恶性风险预测模型并验证其效能。方法回顾性分析2013年1月到2017年7月362例(男194例、女168例,中位年龄61岁)PET/CT示SPN患者的临床和影像学资料,其中恶性291例,良性71例(按病理学或随访结果确诊)。分析良、恶性SPN患者间多种因素差异,通过多因素非条件logistic回归分析筛选危险因素,建立回归方程。以受试者工作特征(ROC)曲线进行模型自身验证,通过k折交叉验证进行模型的组外验证。结果良、恶性结节患者间的年龄及病灶的最大标准摄取值(SUVmax)、大小、分叶、毛刺、胸膜牵拉、血管连接、钙化、空泡、肺气肿等差异具有统计学意义(均P<0.05)。通过多因素logistic回归分析得出,年龄、SUVmax、大小、分叶、钙化、空泡为恶性结节的危险因素,其比值比[OR(95% C7)]分别为:1.040(1.007~1.075)、1.612(1.287-2.017)、1.149(1.074- 1.230)、4.650(2.138- 10.115)、0.216(0. 085~0.548)、3.043(1.302-7.111)。建立模型:P=1/(1+e^-x),x=-5.583 + 0.039×年龄+ 0.477×SUVmax+0.139×大小+1.537×分叶-1.532×钙化+l.H3×空泡。模型曲线下面积[AUC(95% Cl)]为:0.915(0.883-0.947),灵敏度为89.7%、特异性为78.9%;k折交叉验证显示,训练准确性为0.899+0. 011,预测准确性为0.873±0.053。结论患者的年龄及病灶的SUVmax、大小、分叶、钙化和空泡为恶性结节的危险因素;经过验证,该模型具有良好的准确性和稳定性,可辅助临床作出更准确的决策。

Objective To develop a model of malignant risk prediction of solitary pulmonary nodules (SPN) with the metabolic characteristics of the lesion. Methods A total of 362 patients (291 malignant cases and 71 benign cases;194 males, 168 females;median age: 61 years) who underwent PET/CT imaging from January 2013 to July 2017 were analyzed. The diagnosis of malignant SPN was based on pathological results, and that of benign SPN was based on pathological or follow-up results. Differences of clinical/imaging characteristics in patients with benign and those with malignant SPN were analyzed. Risk factors were screened by multivariate non-conditional logistic regression analysis. The self-verification of the model was done by the receiver operating characteristic (ROC) curve analysis, out-of-group verification was performed by k-fold cross-validation. Results There were statistically significant differences in age, maximum standardized uptake value (SUVmax), size, lobulation, spiculation, pleural traction, vessel connection, calcification, vacuole, and emphysema between patients with benign and malignant nodules (all P<0.05). The risk factors for malignant nodules included age, SUVmax, size, lobulation, calcification and vacuole.The odds ratio (OR) values (95% CI) were 1.040(1.007-1.075), 1.612(1.287-2.017), 1.149(1.074-1.230), 4.650(2.138-10.115), 0.216(0.085-0.548), and 3.043(1.302-7.111), respectively. The logistic regression model was as follows: P=1/(1 +e^-x),x=一5.583+0.039×age+0.477×SUVmax+0.139×size+1. 537×lobulation-1.532×calcification+1.113xvacuole. The estimated area under the curve (AUC) for the model was 0.915(95% Cl: 0.883-0.947), sensitivity was 89.7%, specificity was 78.9%. K-fold cross-validation showed that the training accuracy was 0.899±0.011, the predictive accuracy was 0. 873±0.053.Conclusions The risk factors for malignant nodules included age, SUVmax,size, lobulation, calcification and vacuole. After verification, the model has a satisfactory accuracy. It may help clinics make accurate decisions.