Computed tomography radiomics of intratumoral and peritumoral microenvironments for identifying the invasiveness of subcentimeter lung adenocarcinomas.

BACKGROUND

The invasiveness of nodules plays a crucial role in the management and surgical methods selection of lung adenocarcinoma (LAC); however, the ability of traditional chest computed tomography (CT) imaging to detect the invasiveness of subcentimeter LAC is limited.

OBJECTIVE

Development and validation of a model based on computed tomography (CT) radiomics of the intratumoral and peritumoral microenvironments were used to identify the invasiveness of lung adenocarcinomas (LACs) appearing as subcentimeter nodules.

METHODS

A total of 142 consecutive patients with 142 pathologically confirmed subcentimeter LAC nodules were retrospectively studied from January 2020 to December 2023. The demographic data, clinical data, and CT features were retrospectively collected. A total of 2,264 radiomic features were extracted from LAC nodules in the intratumoral and peritumoral microenvironment and then used to construct the radiomic signature with the correlation coefficient and the least absolute shrinkage and selection operator (LASSO) logistic regression and generated radiomic scores (Radscores). A predictive model was constructed based on independent factors selected using a multiple logistic regression model. The performance of the model was evaluated with respect to its discrimination, calibration, and clinical utility.

RESULTS

In a total 142 LAC nodules, including 53 microinvasive adenocarcinoma (MIA) nodules and 89 invasive adenocarcinoma (IAC) nodules, the maximum diameter of nodules in the IAC group was larger than that of the MIA group. The positive rate of the vessel convergence sign (VCS) and vacuole sign in the IAC group were higher than that of the MIA group showing a statistical difference ( < 0.05). Logistic regression analysis showed that the maximum diameters of nodules and VCS were independent factors of IAC, but the predictive model based on CT features (maximum diameter and VCS) had moderate discriminative ability (area under the curve = 0.72), insufficient for standalone clinical use. The Radscores based on gross tumor volume (GTV), gross peritumoral volume (GPTV), and gross peritumoral region (GPR) in the IAC group were significantly higher than those of the MIA group (all  < 0.05, Mann-Whitney U test). The predictive model based on Radscores demonstrated improved discriminative ability (AUCs > 0.75) and calibration compared to CT features, though their clinical utility requires further validation.

CONCLUSIONS

The CT features-based predictive model had limited ability to differentiate the invasiveness in subcentimeter LAC nodules. Models using GTV, GPTV, and GPR Radscores showed improved performance for predicting invasiveness, though further validation is needed, with the GTV-based model performing best. However, this study has limitations, including its retrospective single-center design and potential selection bias due to the small size of subcentimeter lung adenocarcinoma cases.

CLINICAL TRIAL NUMBER

Not applicable.

SUPPLEMENTARY INFORMATION

The online version contains supplementary material available at 10.1186/s12880-025-01882-z.