Joint model based on intratumoral and peritumoral computed tomography radiomics integrated with clinical features for predicting the spread through air spaces in lung adenocarcinoma: a multicenter study

Chen-Zheng-Ren Bao ¹, Rong Zhang ², Sheng-Yao Deng ³

⁰Department of Radiology, Chencun Hospital, Affiliated to Shunde Hospital of Southern Medical University (The First People's Hospital of Shunde), Foshan, China.

Quantitative Imaging in Medicine and Surgery +

|

October 13, 2025

View abstract on PubMed

Summary

A new machine learning model accurately predicts spread through air spaces (STAS) in lung adenocarcinoma using CT scans and clinical data. This tool helps surgeons choose the best treatment, reducing recurrence risk and the need for further surgery.

Area Of Science

Oncology
Radiology
Machine Learning

Background

Spread Through Air Spaces (STAS) is linked to higher recurrence rates after lung cancer surgery.
Accurate STAS assessment requires postoperative pathology, delaying treatment decisions.
Preoperative prediction of STAS can optimize surgical strategy and patient outcomes.

Purpose Of The Study

To develop a joint machine learning model for preoperative STAS prediction in lung adenocarcinoma.
To integrate computed tomography (CT) radiomics (intratumoral and peritumoral) with clinical features.
To facilitate optimal surgical planning and reduce secondary surgeries.

Main Methods

Retrospective study of lung adenocarcinoma patients from three centers.
Development of independent radiomics and clinical feature models using random forest.
Creation of a mixed model combining radiomics and clinical features via logistic regression.
Performance evaluation using AUC, nomogram, calibration, and decision curves.

Main Results

Key predictors of STAS included vacuole sign, crescent sign, gender, air bronchogram, spiculation, and CTR.
The joint model achieved high AUCs: 0.985 (training), 0.988 (internal validation), and 0.851 (external testing).
The model outperformed individual radiomics, clinical models, and other machine learning classifiers (XGBoost, SVM, MLP).

Conclusions

A joint model integrating CT radiomics and clinical data accurately predicts STAS in lung adenocarcinoma.
This preoperative prediction tool aids in selecting appropriate surgical strategies.
The model's nomogram offers an intuitive risk assessment for clinicians, improving treatment precision.

Keywords:

Lung adenocarcinoma peritumoral features prognostic model radiomics spread through air spaces (STAS)