Quantitative CT Analysis Based on Artificial Intelligence in Predicting the Invasiveness of Lung Adenocarcinoma with Ground Glass Nodules Less Than 20 mm
Objective To investigate the value of quantitative parameters CT analysis based on artificial intelligence(AI)in predicting the invasiveness of lung adenocarcinoma with ground glass nodules(GGO)less than 20 mm.Methods The preoperative chest HRCT images of 145 patients with GGNs less than 20 mm conformed adenocarcinoma by surgery and pathology were analyzed retrospectively,including 70 cases of microinvasive adenocarcinoma(MIA)and 75 cases of inva-sive adenocarcinoma(IAC).The CT qualitative parameters of GGN obtained from AI between MIA and IAC groups were compared using independent sample t-test.The independent predictors of IAC were screened by Univariate and binary Lo-gistic regression analysis,and the ROC curve was performed to evaluate the prediction efficacy of these parameters.Results The long diameter,short diameter,mean diameter,volume,CT average,maximum CT value,minimum CT value,media,standard deviation and entropy of GGNs had statistical difference between the two groups(P＜0.05).Logistic regression and the ROC curve analysis showed that the entropy,CT average and volume were the independent risk factors for predicting invasiveness of GGO.Their threshold values were greater than 8.6、-516 HU and 937 mm3,with the corresponding sensi-tivity and specificity were 77.03％,81.30％,73.67％and 74.65％,67.14％,81.71％,respectively.The combination of three parameters had better predictive value than when used alone,with the AUC was 0.918,the sensitivity and specificity were 86.67％and 88.50％.Conclusion Quantitative CT analysis based on AI was helpful to predict the invasiveness of GGO.The combination of the entropy,average CT value and volume had better predictive value than those indexes alone,Which may provide reference for surgery.
NETL
NSTL
万方数据
