A Study Of CT-based Radiomics To Predict Spread Through Air Spaces In Peripheral Non-small Cell Lung Cancer

Objective:To investigate the predictive value of chest CT-based radiomics for spread through air spaces in stage T1 peripheral type lung cancer.Methods:Retrospectively collected 173 patients with surgically pathologically confirmed stage T1 non-small cell lung cancer from January 2020 to December 2021 at Shanxi Bethune Hospital and divided into positive group(n=49 cases)and negative group(n=124 cases)according to the presence or absence of spread through air spaces.The general data and CT morphological features of the two groups of lung cancer patients were compared,and the radiomics analysis of the two groups was performed.Patients were randomly divided into training group(n=122 cases)and validation group(n=51cases)in a 7:3 ratio.3Dslicer software was used to delineate the lesions of the patient’s chest CT images,and the primary area of lung cancer(the main body of the lesion),the peripheral invasion area(the 5mm ring area that expands outward along the edge of the lesion)and the tumor marginal area(the 5mm ring area that retracts inward along the edge of the lesion)were used as the regions of interest.The features were screened by univariate logistic regression and multivariate logistic regression to obtain omics features with statistical differences.Logistic regression was used to establish three radiomics models,namely tumor primary zone model,peripheral invasion region model and tumor marginal region model,and combined with the CT morphological characteristics of lung cancer lesions,three joint models were established,and the performance of each model was evaluated by using the receiver oprerating curve,decision curve analysis and calibration curve,the efficiency of each model was evaluated,and Filter the optimal model.Results:The general data and CT morphological characteristics of lung cancer patients in the negative group and the positive group were compared,and there was no obvious difference in the general data of the patients.In the comparison of the morphological characteristics of lesion CT,there was a statistical difference in the lobar sign(P=0.002),and there was no obvious difference in the other morphological features.The AUC values of the Radiomics model based on the three regions of interest were 0.899,0.825,and0.840 for the training group and 0.876,0.811,and 0.832 for the validation group respectively,The model with the highest AUC value was the primary tumor imaging model(P<0.05),and the AUC values of the combined model established by adding the lobar sign were 0.917,0.835,and 0.851,,respectively,and the AUC values of the validation group were 0.912,0.832,and 0.845.Among them,the combined model of tumor primary region had the highest AUC value(P<0.05),and it was concluded that the Radiomics model and combination model based on tumor primary region had the best predictive effect on lung cancer STAS.Compared with the combined model with lobar sign,there was no significant difference between the radiomics model of the same region of interest(P=0.372 in the primary region of the tumor,P=0.252 in the peripheral invasion region,and P=0.392 in the peripheral invasion region).Conclusion:It is feasible to explore spread through air spaces by CT-based radiomics,and the lobar signs can be used as a risk predictor for STAS of lung cancer.