Objective To establish a diagnostic model based on 18F-FDG PET/CT and clinical data and assess its diagnostic potency for characterizing SPN.Methods From November 2004 to May 2014,164 patients with SPN who underwent 18F-FDG PET/CT scan were retrospectively analyzed.The patients'clinical factors (age,gender,history of smoking and history of malignancy),information on CT (diameter,location and spiculated edge of the lesion) and metabolic information on PET imaging were collected to establish a diagnostic model by using the binary logistic regression.Then,the optimal operating point (OOP)of the established model was set.The diagnostic potencies of the established model and PET were assessed by ROC curve.Results Malignancy was diagnosed in 104 of 164 SPN patients.The rest 60 patients had benign diseases.The factors of age,spiculation(0:no spiculation,1:obvious spiculation) and metabolic information(0:≤ mediastinal blood pool,1:＞mediastinal blood pool) were demonstrated to be useful for the establishment of the model (x2 =5.486,16.240,33.855,all P＜0.05).However,the factors of gender,history of smoking,the diameter and location of lesions showed no influence for the model (x2 =2.452,0.453,0.127,0.390,all P＞0.05) and rejected from the model established.The history of malignancy was excluded from statistical analysis because there were only 2 patients with history of malignancy.The established model was as follows:P=1/(1+e-Z),z=-5.512+0.061xage+2.208xspiculation+3.767×metabolic increase.The ROC AUC of the established model and PET using two-point scoring scale (TPSS) for charactering SPN were 0.92(95％ CI:0.87-0.96)and 0.80(95％ CI:0.73-0.86).The model had higher diagnostic efficacy compared with TPSS (z=4.369,P＜0.05).When P=0.796 7 was set as an OOP,the diagnostic sensitivities of the model and PET for charactering SPN were 91.3％ (95/104) and 94.2％ (98/104) respectively,and no significant difference was found between them (x2 =0.800,P＞0.05).However,significant difference was found between the diagnostic specificities of them (80.0％ (48/60) vs 65.0％ (39/60);x2 =7.111,P＜0.05).Conclusions A new diagnostic model for characterizing SPN based on the information from 18FFDG PET,thin-section CT and clinical data is successfully established.Its sensitivity for diagnosis of lung cancer is high,and its specificity is superior to PET using with TPSS.This model has a potential value for clinical application.%目的 建立1种以18F-FDG PET/CT和临床信息为基础的数学模型,并评价该模型对SPN良恶性的鉴别诊断价值.方法 回顾性分析2004年11月至2014年5月在南方医科大学南方医院PET中心行18F-FDG PET/CT显像的SPN患者164例,纳入患者的临床信息(年龄、性别、吸烟史、恶性肿瘤病史)、薄层CT形态信息(结节大小、位置、毛刺征)和PET代谢信息进行logistic回归分析,建立数学模型,设立最佳工作点,通过ROC曲线分析和评估该模型对SPN的诊断效能.结果 164例SPN患者中,104例经病理证实为肺癌,60例为良性病变.统计学分析结果显示,患者年龄、病灶毛刺征(0:无,1:有)和代谢增高程度(0:不高于纵隔,1:高于纵隔)对诊断模型的建立有价值(x2=5.486、16.240和33.855,均P＜0.05),而性别、吸烟史、病灶大小和位置对诊断模型的建立无价值(x2=2.452、0.453、0.127和0.390,均P＞0.05),恶性肿瘤病史因相关病例数太少(仅2例)而被排除在统计学分析之外.最终得到包含患者年龄、病灶毛刺征和代谢增高程度的SPN诊断模型:P=1/(1+e-z),其中z=-5.512+0.061×年龄+2.208x毛刺征+3.767x代谢增高程度.将该诊断模型与PET二分法进行配对ROC曲线比较,两者的AUC分别为0.92(95％ CI:0.87～ 0.96)和0.80(95％ CI:0.73～0.86),前者的诊断效能较高(z=4.369,P＜0.05).当诊断模型的最佳工作点设为P=0.796 7时,该模型和PET二分法对SPN的诊断灵敏度分别为91.3％(95/104)和94.2％(98/104),差异无统计学意义(x2=0.800,P＞0.05);诊断特异性分别为80.0％ (48/60)和65.0％ (39/60),前者优于后者(x2=7.111,P＜0.05).结论 成功建立以18F-FDG PET、薄层CT和临床信息为基础的SPN鉴别诊断模型;该模型诊断肺癌的灵敏度高,特异性优于传统的PET二分法,具有潜在的临床应用价值.
展开▼
