A computed tomography (CT)-based ternary classification model for predicting the invasiveness of pure ground-glass nodules

Abstract

OBJECTIVE

To develop and validate a multinomial logistic regression model utilising computed tomography (CT) features for the classification of the invasiveness of pulmonary pure ground-glass nodules (pGGNs).

MATERIALS AND METHODS

This retrospective study involved 1572 pathologically confirmed cases of pGGNs, which included atypical adenomatous hyperplasia (AAH), adenocarcinoma in situ (AIS), minimally invasive adenocarcinoma (MIA), and invasive adenocarcinoma (IAC). These cases were categorised into three groups based on invasiveness: precursor glandular lesions (PGLs, AAH + AIS), MIA, and IAC. The cohort was randomly divided into training (70%), testing (15%), and validation (15%) sets using stratified sampling. Univariate and multivariate analyses were conducted to identify candidate predictors, and L1-regularised (Lasso) feature selection was employed to reduce dimensionality. Subsequently, a multinomial logistic regression model was constructed. The model’s performance was evaluated using the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity in both the testing and validation sets.

RESULTS

The mean patient age was 54.9 ± 12.6 years, with 32.4% being male. Seventeen features were identified as significant predictors following Lasso selection. In the testing set, the overall macro-average AUC was 0.793 (95% CI: 0.731-0.853), with a sensitivity of 0.564 and specificity of 0.789. In the validation set, the macro-average AUC was 0.764 (95% CI: 0.696-0.827), with a sensitivity of 0.589 and specificity of 0.803.

CONCLUSION

The proposed CT-based multinomial logistic regression model effectively stratifies pGGNs by invasiveness, providing a noninvasive tool to guide personalised management and enhance preoperative decision-making.