基于CT形态及定量学特征构建多原发肺腺癌、腺体前驱病变风险分层模型

To Constitution a Risk Stratification Model of Multiple Primary Lung Adenocarcinoma and Gland Precursor Lesions Based on CT Morphological and Quantitative Characteristics

目的利用基线CT形态及定量学特征构建多原发肺腺癌(MPLC)、腺体前驱病变不同结节危险度分层的多因素预测模型。方法回顾性分析经手术病理证实的72例MPLC、腺体前驱病变患者临床及影像学资料。根据手术病理类型将不典型腺瘤样增生(AAH)及原位癌(AIS)归为低危组,而将微浸润腺癌(MIA)及浸润性腺癌(IAC)归为高危组。结节形态学特征[成分、边缘(毛刺和分叶)、空腔(空泡、扩张支气管征)、胸膜牵拉]由人工智能(AI)肺结节分析软件自动获得,然后由1名具有5年胸部CT经验的影像科医师再次评估。结节定量CT参数(最大径、长短径之比、平均CT值及体积)由后处理工作站(Lung VCAR,GE)分析获得。采用SPSS 22.0软件进行统计分析,Logistic回归分析筛选预测因子,构建预测模型,受试者工作特征曲线(ROC)分析评价模型效能。P<0.05为差异有统计学意义。结果72例患者经手术病理证实160枚结节,其中双原发结节57例(79.17%),女性52例(72.22%)。高危组、低危组结节最大径分别为(13.85±6.14)mm、(6.10±2.45)mm,长短径之比分别为1.38±0.25、1.30±0.25,平均CT值分别为(-532.71±175.88)HU、(-669.23±103.92)HU,体积分别为(1618.00±2149.79)mm^(3)、(178.75±198.02)mm^(3)。高危组、低危组结节最大径、平均CT值和结节体积间差异均有统计学意义(P=0.000,0.001,0.000)。结节最大径预测高危组的最佳截断值为7 mm,曲线下面积(AUC)为0.910,敏感度和特异度分别为81.25%和86.61%。结节体积预测高危组的最佳截断值为323 mm^(3),AUC值、敏感度、特异度分别为0.904、91.67%、77.68%。将定量参数结节最大径、体积、平均CT值纳入Logistic回归分析,构建组合变量。ROC曲线分析显示,组合变量预测高危组结节的AUC值为0.939,敏感度90.00%,特异度85.71%,差异具有统计学意义(P<0.0001)。形态学特征上,结节内部是否含有实性成分、实性成分多少[纯磨玻璃密度结节(pGGN)、混合磨玻璃密度结节(mGGN)和实性结节(SN)]在高、低危组间有显著性差异(P=0.000、0.000、0.016)。毛刺、分叶、胸膜牵拉、空泡、扩张支气管征等形态学特征在两组间差异同样具有统计学意义(P<0.05)。结论CT形态学特征、定量参数能够在一定程度上对MPLC及腺体前驱病变进行危险度分层,多参数组合定量模型的预测效能最佳。

Objective To constitute a multi-parameter risk stratification predictive model based on the morphological and quantitative features of baseline CT to distinguish the multiple primary lung cancer(MPLC)and glandular prodrome lesions.Methods The clinical and imaging data of enrolled 72 patients confirmed by operative pathology were retrospectively analyzed.According to the pathological classification,atypical adenomatous hyperplasia(AAH)and adenocarcinoma in situ(AIS)were classified into the low risk group,while minimally invasive adenocarcinoma(MIA)and invasive adenocarcinoma(IAC)were classified into the high risk group.The morphological features(composition,margin[spiculation and lobulation],airspace[vacuole,air bronchogram],pleural retraction)of the different MPLC nodules were automatically achieved by the artificial intelligence(AI)pulmonary nodule analysis software,then reevaluated by one radiologist with 5 years experience in chest CT imaging.The quantitative parameters of CT(maximum diameter,ratio of long to short diameter,volume and mean CT value)were measured in the post processing workstation(Lung VCAR software,GE).SPSS 22.0 software was used to analyze.Logistic regression analysis was used to generate a combined risk model for predicting high risk lesions.ROC analysis was used to evaluate the efficacy of the model.P values less than 0.05 were considered statistically significant.Results 160 nodules were confirmed by surgery and pathology in 72 patients,including 57 cases(79.17%)of dual primary lesions and 52 cases(72.22%)of females.The maximum diameter of lesion and the ratio of long to short diameter of lesion in high and low risk group were(13.85±6.14)mm,(6.10±2.45)mm and 1.38±0.25,1.30±0.25,respectively.The mean CT values were(-532.71±175.88)HU for high-risk group,and(-669.23±103.92)HU for low-risk group.And,the volumes of lesions were(1618.00±2149.79)mm~3 and(178.75±198.02)mm~3 for two groups.The differences in maximum diameter,mean CT value and the volume of lesion between high and low risk group were statistically significant(P=0.000,0.001,0.000).For the maximum diameter of lesions,the best cut-off value for predicting high risk group lesions was 7 mm,AUC value was 0.910,and the sensitivity and specificity were 81.25%and 86.61%,respectively.The best cut-off value for volume to predict high risk group lesions was 323 mm3,and the AUC value,sensitivity and specificity were 0.904,91.67%and 77.68%,respectively.The maximum diameter,volume and average CT value of lesions were included in logistic regression analysis to construct a combined variable.ROC curve analysis showed that the AUC value of combined variable in predicting high risk group lesions was 0.939,the sensitivity was 90.00%,and the specificity was 85.71%.The difference was statistically significant(P<0.0001).For morphological features of lesions,the presence of solid components(pure ground glass nodule[pGGN]mixed with ground glass nodule[mGGN]and solid nodule[SN])was significantly different between high and low risk groups(P=0.000,0.000,0.016).The spiculation,lobulation,pleural retraction,vacuole and air bronchogram were also significantly different between two groups with P value less than 0.05.Conclusion CT morphological characteristics and quantitative parameters might be used for the risk stratification of MPLC and gland precursor lesions to some extent,and the combined multi-parameters quantitative model showed the best predictive performance.