Prediction Of Lung Adenocarcinoma Growth Pattern Based On Lung Nodules CT Image Features And Quantitative Features Of Artificial Intelligence Software

Objective: Based on lung nodules CT image features and the quantitative features of Artificial Intelligence(AI)software,a model for predicting the growth pattern of lung adenocarcinoma was established,and its application value in differentiating the growth pattern of lung adenocarcinoma with lepidic growth pattern from non-lepidic growth pattern was discussed.Method: A retrospective analysis was performed on clinical and imaging data of 191 cases(204GGNs)of lung adenocarcinoma confirmed by surgery and pathology in the Fourth Hospital of Hebei Medical University from January2020 to December 2020,with preoperative CT images showing as ground-glass nodule(GGN).The age,gender,and smoking history of the patients were recorded,and the image characteristics of each GGN were analyzed,including location,nodule type,tumor-lung interface,deep lobular sign,burr sign,vacuole sign,internal vascular sign,bronchial anomaly sign,and pleural pulling sign,United Imaging intelligent lung nodule CT image-assisted examination software(u AI-Chest Care)was used to measure quantitative characteristics of each GGN: the average diameter,total volume,total mass,volume proportion and mass proportion of solid components(for voxel with CT value ≥-50HU),as well as the maximum,minimum,mean,median,standard deviation,skewness,kurtosis,and entropy of each GGN CT value.According to postoperative pathological results,the enrolled GGNs were divided into lung adenocarcinoma with lepidic growth pattern(136GGNs)and lung adenocarcinoma with nonlepidic growth pattern(68GGNs)according to growth pattern.Univariate and multivariate logistic regression analysis was used to screen the CT image features and AI quantitative features highly correlated with the growth pattern of lung adenocarcinoma,then a model for predicting growth pattern of lung adenocarcinoma was established.Receiver Operating Characteristic(ROC)Curve was used to evaluate the diagnostic efficiency of the model,and the Area Under the Curve(AUC)of the model was calculated,Delong test was used to compare AUCs of independent predictors and combined models.Results: Univariate analysis showed that sex,age,smoking history,nodule type,deep lobulation sign,burr sign,bronchial anomaly sign,pleural drawing sign,mean diameter,total volume,total mass,volume proportion of solid component,mass proportion of solid component,maximum,minimum,median,standard deviation,skewness,kurtosis and entropy of CT value were correlated with the growth pattern of lung adenocarcinoma(all P < 0.05),Multivariate logistic regression analysis showed that bronchial abnormalities,mean diameter,volume ratio of solid component and mass ratio of solid component were independent predictors for differentiating the growth pattern of lung adenocarcinoma.Therefore,a combined model for predicting the growth of lung adenocarcinoma was established,the AUC of which was 0.871,the sensitivity was 79.4%,the specificity was 83.1%,and the accuracy was 84.8%.Diagnostic efficacy of this model was higher than that of a single independent predictor(AUCs of bronchial anomaly sign,mean diameter,solid component volume ratio,and solid component mass ratio were 0.728,0.831,0.780,and0.779,respectively).Conclusion: The model for predicting the growth pattern of lung adenocarcinoma established based on lung nodules CT image features and the quantitative features of AI software has good diagnostic efficacy,and can provide a basis for clinical treatment decision-making.