基于人工智能分析软件构建原发肺腺癌PD-L1表达的分层预测模型

To Constitute A Stratified Prediction Model of PD-L1 Expression in Primary Lung Adenocarcinoma Based on Artificial Intelligence Analysis Software

目的利用人工智能分析软件获取的影像学特征构建原发肺腺癌PD-L1表达的分层预测模型。方法回顾性分析2020年6月至2021年6月共129例经手术证实为原发肺腺癌的患者影像资料,由人工智能肺结节分析软件自动获得病灶影像表征、CT定量参数及纹理特征,再由1名具有5年胸部CT诊断经验的影像科医师进行复核,有争议时由一名具有25年诊断经验的医师最终阅片决定。采用免疫组织化学染色法检测患者病理标本中PD-L1的表达,将患者分为PD-L1阳性表达组(TPS≥1%)、PD-L1阴性表达组(TPS<1%)。采用SPSS 26.0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PD-L1阳性表达组实性体积、实性体积占比均明显高于阴性表达组,P值均<0.001。两组间实性质量分别为5.57、1.59mg×10^(3),差异具有明显统计学意义(P=0.000),阳性表达组实性质量占比也明显高于阴性组,分别为0.73、0.41(P=0.000)。阳性组、阴性组总质量分别为6.40、2.76mg×10^(3)(P=0.015)。PD-L1阳性表达组CT最大值、平均值均高于阴性表达组(P值均=0.000)。两组间偏度及熵值差异有统计学意义(P=0.001、0.002)。将实性成分、CT值、纹理特征纳入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0.887,敏感度74.5%,特异度90.2%,差异具有统计学意义(P<0.001)。结论基于人工智能分析软件的影像学定量特征能够用于预测原发肺腺癌PD-L1表达水平,多参数组合变量的预测模型最佳。

Objective To constitute a stratified prediction model of PD-L1 expression in primary lung adenocarcinoma based on imaging features obtained by artificial intelligence analysis software.Methods The imaging data of 129 patients with primary lung adenocarcinoma confirmed by operation from June 2020 to June 2021 were analyzed retrospectively.The imaging features of the lesions and CT quantitative parameters and texture features were automatically obtained by the artificial intelligence lung nodule analysis software.It was then reviewed by a radiologist with 5 years experience in the diagnosis of chest CT,and the final decision was made by a radiologist with 25 years'experience in diagnosis.The expression of PD-L1 in pathological specimens was detected by immunohistochemical staining.The patients were divided into PD-L1 positive expression group(TPS≥1%)and PD-L1 negative expression group(TPS<1%).SPSS26.0 software was used for statistical analysis.Logistic regression analysis was used to screen the predictive factors and construct predictive model.ROC analysis was used to evaluate the efficacy of the model.P values less than 0.05 were considered statistically significant.Results The volume and proportion of solid components in PD-L1 positive expression group were significantly higher than those in negative expression group(P<0.001).The mass of solid components between the two groups was 5.57 and 1.59mg×10^(3)respectively.The difference was statistically significant(P=0.000).The proportion of solid components in the positive expression group was also significantly higher than that in the negative group,which were 0.73 and 0.41 respectively(P=0.000).The total mass of positive group and negative group was 6.40,2.76 mg×10^(3)respectively.The maximum and average CT values in PD-L1 positive expression group were higher than those in negative expression group(P=0.000).There were significant differences in skewness and entropy between the two groups(P=0.001,0.002).The solid components,CT values and texture features were included in the Logistic regression analysis to constitute a prediction model.The ROC curve analysis showed that the AUC value of the combined variables was 0.887,the sensitivity was 74.5%,and the specificity was 90.2%.The difference was statistically significant(P<0.001).Conclusion The imaging quantitative features based on artificial intelligence analysis software can be used to predict the expression level of PD-L1 in primary lung adenocarcinoma,and the prediction model of multi-parameter combination variables is the best.