长期吸烟者合并肺纤维化的孤立性肺结节^(18)F-FDG PET/CT恶性风险预测模型的建立

Prediction model for probability of malignancy in solitary pulmonary nodules on ^(18)F-FDG PET/CT of smokers with pulmonary fibrosis

目的构建并验证长期吸烟者肺间质纤维化背景下的孤立性肺结节(SPNs)18F-脱氧葡萄糖(FDG)PET/CT恶性风险预测模型。方法收集2011年1月至2019年12月间于青岛大学附属医院PET/CT中心行18F-FDG PET/CT显像,且有明确吸烟史、同机CT示有肺间质纤维化合并SPNs的169例患者进行回顾性分析,均为男性,年龄68(63,75)岁。以病理学诊断结果或肺内病灶随访的影像学资料(随访时间≥2年)为标准,判断结节良恶性;运用χ^(2)检验、Mann-Whitney U检验比较良恶性病灶的临床特征(年龄、吸烟指数)、形态特征(病灶最大径、密度、位置、分布、与纤维化区域相对位置、毛刺征、分叶征、钙化、空泡征、血管集束征、胸膜凹陷征、肺气肿及双肺纤维化严重程度)和代谢特征[病灶最大标准摄取值(SUVmax)],将具有统计学意义的差异变量纳入多因素logistic回归,筛选结节恶性的独立危险因素并建立风险预测模型。以受试者工作特征(ROC)曲线的曲线下面积(AUC)及k折交叉验证(k=10)验证模型。结果共发现SPNs 222个,其中恶性157个、良性65个。单因素分析显示,吸烟指数,结节是否伴毛刺征、分叶征、血管集束征、钙化、肺气肿,结节大小,与纤维化区域的相对位置,SUV_(max),双肺纤维化严重程度在良恶性结节中差异均有统计学意义(z值:2.514~9.858,χ^(2)值:4.353~18.442,均P<0.05)。多因素logistic回归分析显示,钙化、血管集束征及SUV_(max)为肺间质纤维化背景下恶性结节的独立危险因素[比值比(OR):0.048~2.534,均P<0.05],据此构建的预测模型为:恶性概率P=1/(1+e-x),x=-1.839-3.033×钙化+0.930×血管集束征+0.754×SUV_(max)(结节具有钙化或血管集束征赋值为1,否则赋值为0)。自身验证ROC曲线下面积为0.932(95%CI:0.895~0.969),模型灵敏度、特异性分别为87.9%、86.2%。k折交叉验证示,测试组预测准确性为0.847±0.075,训练组预测准确性为0.862±0.010。结论钙化、血管集束征和SUVmax是长期吸烟者肺间质纤维化背景下恶性SPNs的独立危险因素,基于上述指标的模型判断恶性SPNs具有较高的诊断效能。

Objective To establish and validate a malignant risk prediction model of solitary pulmonary nodules(SPNs)with pulmonary fibrosis in long-term smokers based on 18F-flurodeoxyglucose(FDG)PET/CT.Methods PET/CT images of 222 SPNs combined with pulmonary fibrosis which were shown in integrated CT scan in 169 patients(all males;age 68(63,75)years)were analyzed retrospectively.All patients were examined in PET/CT Center of the Affiliated Hospital of Qingdao University from January 2011 to December 2019 and all had definite smoking history.The benign and malignant nodules were judged according to the pathological diagnosis or follow-up imaging data of lung lesions(follow-up≥2 years).The clinical characteristics(age,smoking index),morphological characteristics(longest diameter of lesion,density,location,distribution,relative position of fibrosis,spiculation,lobulation,calcification,vacuole,vascular convergence,pleural indentation,emphysema and severity of bilateral pulmonary fibrosis)and metabolic characteristics(maximum standardized uptake value(SUVmax))of the benign and malignant lesions were analyzed byχ^(2) test and Mann-Whitney U test.Then multivariate logistic regression analysis was applied to select independent risk factors of malignant nodules,and a risk prediction model was established and verified by the area under the receiver operating characteristic(ROC)curve and k-fold cross validation(k=10)respectively.Results Among 169 patients,222 SPNs were detected(157 malignant nodules,65 benign nodules).Univariate analysis showed that smoking index,speculation,lobulation,vascular convergence sign,calcification,emphysema,nodule size,relative position of nodule and fibrosis,SUVmax and severity of bilateral pulmonary fibrosis were significantly different between the benign and malignant nodules(z values:2.514-9.858,χ^(2) values:4.353-18.442,all P<0.05).Result of multivariate logistic regression analysis showed that calcification,vascular convergence and SUVmax were the independent risk factors of malignant nodules combined with pulmonary fibrosis(odds ratio(OR):0.048-2.534,all P<0.05).The risk prediction model was as follow:P=1/(1+e-x),x=-1.839-3.033×calcification+0.930×vascular convergence+0.754×SUVmax(with calcification/vascular convergence=1,without calcification/vascular convergence=0).The area under ROC curve was 0.932(95%CI:0.895-0.969),and the sensitivity and specificity of the model were 87.9%and 86.2%,respectively.Results of k-fold cross validation showed that the prediction accuracy of 10 test sets was 0.847±0.075,and was 0.862±0.010 in training sets.Conclusions Calcification,vascular convergence and SUVmax are independent risk factors of malignant SPNs combined with pulmonary fibrosis in long-term asymptomatic smokers.The model based on the above variables presents high diagnostic efficiency in diagnosing malignant SPNs.