The Predictive Value Of Artificial Intelligence-Based CT Quantitative Parameters And Lung-RADS Grading For The Invasiveness Of Ground-Glass Nodule Type Lung Adenocarcinoma

Objective The invasiveness of ground-glass nodules(GGNs)in lung adenocarcinoma was predicted by artificial intelligence(AI)-derived CT parameters and Lung-RADS classification.Methods Data from patients who had an operation for early-stage lung adenocarcinoma at General Hospital of Ningxia Medical University between January 2017 and November 2020 were collected by us.The patients presented with GGNs on preoperative thin-section chest CT and were confirmed to have lung adenocarcinoma by paraffin pathology after surgery(including AAH,AIS,MIA,and IAC).The overall sample included 208 patients with 208 GGNs,which were classified into an IAC group(88 cases)and a non-IAC group(120 cases;including AAH,AIS,and MIA)based on postoperative pathological diagnosis.AI software was used to automatically detect CT quantitative and qualitative parameters as well as LungRADS classification based on preoperative chest CT.The relationship between CT parameters/Lung-RADS classification and nodule invasiveness was analyzed.Additionally,a7:3 ratio was used to split all samples into training and validation groups.Two prediction models for invasive adenocarcinoma were constructed based on the training data,and their discrimination,calibration,and effectiveness were systematically evaluated.Model 2 was ultimately chosen for visualization using a column chart.Results 1.The difference analysis between the IAC and non-IAC groups showed that except for gender and nodule location,there were statistically significant differences in nodule type,diameter,volume,mass,CT value,and Lung-RADS classification(p<0.05),indicating that these variables were related to GGN invasiveness.2.Univariate logistic regression analysis revealed that age,nodule type,mass-like subsolid nodule(MASS),effective diameter,maximum diameter,minimum diameter,nodule volume,mean CT value,maximum CT value,minimum CT value and Lung-RADS classification were correlated with invasive adenocarcinoma(IAC)and could be used as predictive factors for IAC.3.Model 1 for predicting the probability of IAC in GGNs was constructed based on CT parameters including mass,maximum diameter,mean CT value,minimum CT value,and LungRADS classification,while Model 2 was constructed by removing Lung-RADS classification.The comparison between Models 1 and 2 showed that in the training group,the area under the curve(AUC)of the prediction probability curve for Model 1 was 0.859,while that for Model 2was 0.776.In the validation group,the AUC for Model 1 was 0.769,while that for Model 2 was0.776.The Delong test indicated that Models 1 and 2 had no significant difference in AUC(p>0.05).Moreover,NRI,IDI,DCA,and calibration curves also revealed no significant difference between the two models,indicating that both models had excellent predictive ability.Conclusion AI-derived CT quantitative and qualitative features were found to be related to GGN invasiveness.Moreover,the GGN probability prediction model for IAC constructed based on these features showed good performance,providing a new approach for early screening and diagnosis of lung adenocarcinoma with GGNs.