【摘要】 目的 探讨基于CT特征与人工智能(AI)定量参数构建的模型在预测长径≤2 cm肿瘤性混合磨玻璃结节(mGGN)浸润性中的价值。方法 回顾性选取254例经病理证实且长径≤2 cm的肿瘤性mGGN患者,将其分为非浸润组[包括28例原位腺癌(AIS)和112例微浸润腺癌(MIA)]和浸润组[114例浸润性腺癌(IAC)],并以7:3的比例随机分为训练集(n=177)和验证集(n=77)。采用逐步回归法对单因素logistic回归分析中有统计学意义的变量进行筛选,确定最佳变量组合并构建CT特征模型、AI定量模型和综合模型。结果 年龄、异常支气管征、深分叶征和网状征是构建CT特征模型的最佳变量组合,三维长径和偏度是构建AI定量模型的最佳变量组合。CT特征模型、AI定量模型与综合模型在训练集和验证集中的曲线下面积(AUC)分别为0.859、0.883、0.904和0.773、0.843、0.847。在训练集和验证集中,综合模型的预测效能均优于CT特征模型(P<0.05),而与AI定量模型之间无统计学差异(P>0.05)。决策曲线分析显示综合模型比CT特征模型和AI定量模型具有更好的净获益。结论基于CT特征与AI定量参数构建的综合模型有助于量化长径≤2 cm肿瘤性mGGN的浸润性,其净获益优于单一模型。更多还原

【Abstract】 Objective To explore the value of the model constructed based on CT features and artificial intelligence(AI) quantitative parameters in predicting the invasiveness of neoplastic mixed ground-glass nodule(mGGN) with a long axis diameter≤2 cm.Methods A retrospective selection was made of 254 pathologically confirmed neoplastic mGGN patients with a long axis diameter ≤2 cm.All patients were categorized into the non-invasive group[including 28 cases of adenocarcinoma in situ(AIS) and 112 cases of minimally invasive adenocarcinoma(MIA)] and the invasive group [114 cases of invasive adenocarcinoma(IAC)],and were randomly divided into the training set(n=177) and the validation set(n=77) at a ratio of 7:3.The stepwise regression method was applied to screen the statistically significant variables in the univariate logistic regresssion analysis to determine the optimal variable combinations,and to construct the CT feature model,AI quantitative model and comprehensive model.Results Age,abnormal bronchial sign,deep lobulation sign and reticular sign were the optimal variable combinations for constructing CT feature model,while three-dimensional long axis diameter and skewness were the optimal variable combinations for constructing AI quantitative model.The area under the curve(AUC) of the CT feature model,AI quantitative model and comprehensive model in the training set and validation set were0.859,0.883,0.904 and 0.773,0.843 and 0.847,respectively.In both the training set and the validation set,the predictive performance of the comprehensive model was superior to that of the CT feature model(P<0.05),while there was no significant differences compared with the AI quantitative model(P>0.05).The decision curve analysis demonstrated that the comprehensive model had superior net benefit than the CT feature model and AI quantitative model.Conclusion The comprehensive model constructed based on CT features and AI quantitative parameters facilitates invasiveness quantification of neoplastic mGGN with a long axis diameter ≤2 cm,demonstrating superior net benefit over single model.更多还原