人工智能量化参数对亚实性结节肺腺癌浸润程度的预测价值

Predictive value of quantitative parameters by artificial intelligence for the infiltration degree of lung adenocarcinoma with subsolid nodules

目的:探讨人工智能(AI)肺结节定量参数预测亚实性结节(SSN)肺腺癌浸润程度的临床价值。方法:回顾性分析经手术病理证实的89例(共94个结节)SSN肺腺癌患者的临床及胸部CT资料。根据病理结果,将其分为非浸润性病变组:原位腺癌(AIS)6个、微浸润性腺癌(MIA)29个;浸润性病变组:浸润性腺癌(IAC)59个。比较2组临床资料及结节AI定量参数间的差异,采用单因素与多因素二元logistic回归分析筛选SSN肺腺癌浸润程度的独立因素并建立预测模型,以ROC曲线分析模型预测概率和独立预测因子对肺腺癌浸润程度的预测价值。结果:2组间年龄、长径、短径、恶性概率、体积、质量、最大CT值、平均CT值、CT值方差、球型度、最大面面积、表面积、3D长径、长短径平均值、紧凑度差异均有统计学意义(均P<0.05)。logistic回归分析显示,质量(OR=1.002,P=0.005)和平均CT值(OR=1.006,P=0.001)是SSN肺腺癌浸润程度的独立预测因子,诊断阈值分别为202.2 mg、-463.5 HU。预测模型为logit(P)=0.002X_(1)+0.006X_(2)+2.554。结论:AI量化参数对预测SSN肺腺癌浸润程度有一定临床参考价值,质量和CT平均值可作为诊断浸润程度的可靠指标。模型预测概率较质量、CT平均值更准确预测IAC的发生。

Objective:To investigate the clinical value of quantitative parameters by artificial intelligence(AI)in predicting the infiltration degree of lung adenocarcinoma with subsolid nodules(SSNs).Methods:The clinical and chest CT data of 89 patients(94 SSNs)with lung adenocarcinoma confirmed by surgery and pathology were retrospectively analyzed.According to the pathological results,94 SNNs were divided into two groups:the noninvasive group,including 6 SSNs of adenocarcinoma in situ(AIS)and 29 SSNs of minimally invasive adenocarcinoma(MIA),and the invasive group,including 59 SSNs of invasive adenocarcinoma(IAC).The differences in clinical data and AI quantitative parameters between the two groups were compared.Univariate and multivariate binary logistic regression analysis were used to screen independent factors and establish a prediction model for the infiltration degree of lung adenocarcinoma with SSNs.ROC curves were used to analyze the prediction probability of the model and the predictive value of independent predictors for the infiltration degree of lung adenocarcinoma.Results:There were significant differences in age,the maximum and minimum diameter,probability of malignancy,volume,mass,the maximum and mean CT value,CT value variance,sphericity,the maximum surface area,the surface area,the 3D diameter,mean diameter and compactness between the two groups(all P<0.05).Logistic regression analysis showed that the mass(OR=1.002,P=0.005)and the mean CT value(OR=1.006,P=0.001)were the independent predictors of the infiltration degree of lung adenocarcinoma of SSNs,and the diagnostic thresholds were 202.2 mg and-463.5 HU,respectively.The predictive model was logit(P)=0.002X_(1)+0.006X_(2)+2.554.Conclusions:AI quantitative parameters have a certain clinical reference value for predicting the infiltration degree of lung adenocarcinoma with SSNs.The mass and the mean CT value can be used as the reliable indicators to diagnose the infiltration degree.The prediction probability of the model is more accurate in predicting the occurrence of IAC than the mass and the mean CT value.