基于人工智能定量模型对磨玻璃结节Ⅰ期浸润性肺腺癌病理亚型的预测价值  

Predictive value of the quantitative model based on artificial intelligence for pathological subtypes of stageⅠinvasive lung adenocarcinoma with ground glass nodule

作　　者：邓琦[1] 徐志锋[1] 成东亮 周涛[1] 李勤祥[1] DENG Qi;XU Zhifeng;CHENG Dongliang;ZHOU Tao;LI Qinxiang(Department of Medical Imaging,Foshan First People’s Hospital,Foshan,Guangdong Province 528000,China)

机构地区：[1]佛山市第一人民医院医学影像科,广东佛山528000

出　　处：《实用放射学杂志》2023年第12期1941-1944,2000,共5页Journal of Practical Radiology

基　　金：佛山市“十四五”医学重点专科和培育专科项目(FSGSP145036)。

摘　　要：目的探讨人工智能(AI)定量模型对磨玻璃结节(GGN)Ⅰ期浸润性肺腺癌病理亚型的预测价值。方法回顾性分析经手术病理证实为Ⅰ期浸润性肺腺癌的GGN患者118例(124个病灶),根据病理亚型分为贴壁为主型腺癌(LPA)组(46个病灶)和非贴壁为主型腺癌(n-LPA)组(78个病灶)。记录相关AI定量参数,包括最长径、总体积、实性体积占比、总质量、实性质量占比、最大CT值、最小CT值、平均CT值。采用单因素及多因素logistic回归分析筛选出n-LPA的独立预测因子,绘制Nomogram图量化独立风险因素,利用受试者工作特征(ROC)曲线评价模型诊断效能。结果二元logistic回归分析表明,实性质量占比[比值比(OR)=1.965,95%置信区间(CI)1.515~2.549]和平均CT值(OR=1.020,95%CI 1.004~1.036)是判断n-LPA的独立预测因子(P<0.05)。Nomogram图量化独立风险因素显示以上预测模型与实际结果的一致性良好,一致性指数(C指数)值为0.872(95%CI 0.791~0.953)。ROC曲线分析表明,联合上述2个独立预测指标的鉴别诊断效能[曲线下面积(AUC)=0.829]优于单独指标中的实性质量占比(AUC=0.788)与平均CT值(AUC=0.765),相应的敏感度和特异度分别为87.2%、84.8%,与病理结果的一致性尚可(Kappa=0.667)。结论AI定量模型中的实性质量占比、平均CT值可有效协助预测GGNⅠ期浸润性肺腺癌的病理亚型,联合上述2个指标能提高CT对LPA与n-LPA的鉴别诊断效能。Objective To explore the predictive value of artificial intelligence(AI)quantitative model for pathological subtypes of stageⅠinvasive lung adenocarcinoma with ground glass nodule(GGN).Methods A total of 118 cases(124 lesions)of GGN patients with stageⅠinvasive lung adenocarcinoma confirmed by surgery and pathology were analyzed retrospectively,and they were divided into lepidic predominant adenocarcinoma(LPA)group(46 lesions)and non-lepidic predominant adenocarcinoma(n-LPA)group(78 lesions)according to the pathological subtype results.Some relevant AI quantitative parameters were recorded,including the longest diameter,total volume,the percentage of solid volume,total mass,the percentage of solid mass,maximum CT value,minimum CT value,and average CT value.The independent predictors of n-LPA were screened by univariate and multivariate logistic regression analysis,the independent risk factors were quantified by Nomogram,and the diagnostic efficiency of the model was evaluated by using receiver operating characteristic(ROC)curve.Results Binomial logistic regression analysis showed that the percentage of solid mass[odds ratio(OR)=1.965,95%confidence interval(CI)1.515-2.549]and average CT value(OR=1.020,95%CI 1.004-1.036)were independent predictors of n-LPA(P<0.05).The Nomogram to quantify the independent risk factors showed that the above prediction model was in good agreement with the actual results,and the C-index value was 0.872(95%CI 0.791-0.953).ROC curve analysis showed that the diagnostic performance of the combination of the above two indexes[area under the curve(AUC)=0.829]was better than that of the solid mass percentage(AUC=0.788)and the average CT value(AUC=0.765)of the single indexes,and the corresponding sensitivity and specificity were 87.2%and 84.8%,respectively,which were consistent with the pathological results(Kappa=0.667).Conclusion The percentage of solid mass and the average CT value in the AI quantitative model can effectively help predict the pathological subtypes of GGN stageⅠinvasive

关 键 词：人工智能 磨玻璃结节 肺腺癌 病理亚型 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] R734.2[自动化与计算机技术—控制科学与工程] R446.8[医药卫生—肿瘤]