Objectives To investigate the value of machine learning classifiers incorporating dual-layer spectral CT (DLCT) parameters for preoperative prediction of spread through air spaces (STAS) in patients with lung adenocarcinoma. Materials and methods This two-center retrospective study included 246 lung adenocarcinoma patients from center I (training cohort) and 193 patients from center II (test cohort). DLCT parameters and clinicoradiologic characteristics were collected. Univariable and multivariable logistic regression analyses were performed to identify independent factors for STAS, among which DLCT parameters were used to develop DLCT-based models (Model-DLCT) using five machine learning classifiers. Similarly, clinicoradiologic characteristics associated with STAS were subsequently combined with DLCT parameters to develop combined models (Model-COM). The prediction performances were evaluated using the receiver operating characteristic curve and decision curve analysis (DCA). Results Venous phase electron density (odds ratio (OR) = 1.042, p = 0.02) and venous phase normalized iodine concentration (OR = 73.015, p < 0.01) were used to construct the Model-DLCT. Among the five classifiers, the extreme gradient boosting (XGBoost)-based Model-DLCT achieved the best performance, with AUC values of 0.833 [95% CI: 0.777–0.889] and 0.829 [95% CI: 0.773–0.886] in the training and test cohorts, respectively. The consolidation/tumor ratio (CTR, OR = 17.865, p = 0.01) was the only significant clinicoradiologic predictor. The combined model, integrating CTR with the Model-DLCT, demonstrated modestly improved discrimination with AUCs of 0.862 (95% CI: 0.812–0.911) and 0.832 (95% CI: 0.774–0.889) in the two cohorts. DCA further confirmed its clinical utility. Conclusion A machine learning model integrating DLCT quantitative parameters with clinicoradiologic characteristics provides a robust tool for preoperative prediction of STAS in patients with lung adenocarcinoma. Key Points <Emphasis Type="BoldItalic">Question</Emphasis> Does the application of dual-layer spectral CT improve the prediction of spread through air spaces in lung adenocarcinoma? <Emphasis Type="BoldItalic">Findings</Emphasis> A machine learning model, integrating dual-layer spectral CT parameters with clinicoradiologic features, demonstrated optimal prediction of spread through air spaces. <Emphasis Type="BoldItalic">Clinical relevance</Emphasis> Dual-layer spectral CT demonstrates significant value in predicting spread through air spaces in lung adenocarcinoma and could inform preoperative risk stratification and surgical decision-making. Graphical Abstract

Dual-layer spectral CT for predicting spread through air spaces in lung adenocarcinoma: a dual-center study

期刊论文
发表日期：

2026年4月4日

查看全文

Changxing Fang,
Yanfen Cui,
Fanger Li,
Xiang Li,
Ying Qiao,
Xiaotang Yang

摘要

Objectives

To investigate the value of machine learning classifiers incorporating dual-layer spectral CT (DLCT) parameters for preoperative prediction of spread through air spaces (STAS) in patients with lung adenocarcinoma.

Materials and methods

This two-center retrospective study included 246 lung adenocarcinoma patients from center I (training cohort) and 193 patients from center II (test cohort). DLCT parameters and clinicoradiologic characteristics were collected. Univariable and multivariable logistic regression analyses were performed to identify independent factors for STAS, among which DLCT parameters were used to develop DLCT-based models (Model-DLCT) using five machine learning classifiers. Similarly, clinicoradiologic characteristics associated with STAS were subsequently combined with DLCT parameters to develop combined models (Model-COM). The prediction performances were evaluated using the receiver operating characteristic curve and decision curve analysis (DCA).

Results

Venous phase electron density (odds ratio (OR) = 1.042, p = 0.02) and venous phase normalized iodine concentration (OR = 73.015, p < 0.01) were used to construct the Model-DLCT. Among the five classifiers, the extreme gradient boosting (XGBoost)-based Model-DLCT achieved the best performance, with AUC values of 0.833 [95% CI: 0.777–0.889] and 0.829 [95% CI: 0.773–0.886] in the training and test cohorts, respectively. The consolidation/tumor ratio (CTR, OR = 17.865, p = 0.01) was the only significant clinicoradiologic predictor. The combined model, integrating CTR with the Model-DLCT, demonstrated modestly improved discrimination with AUCs of 0.862 (95% CI: 0.812–0.911) and 0.832 (95% CI: 0.774–0.889) in the two cohorts. DCA further confirmed its clinical utility.

Conclusion

A machine learning model integrating DLCT quantitative parameters with clinicoradiologic characteristics provides a robust tool for preoperative prediction of STAS in patients with lung adenocarcinoma.

Key Points

Question Does the application of dual-layer spectral CT improve the prediction of spread through air spaces in lung adenocarcinoma?

Findings A machine learning model, integrating dual-layer spectral CT parameters with clinicoradiologic features, demonstrated optimal prediction of spread through air spaces.

Clinical relevance Dual-layer spectral CT demonstrates significant value in predicting spread through air spaces in lung adenocarcinoma and could inform preoperative risk stratification and surgical decision-making.

Graphical Abstract