A key pattern of lung cancer spread. Spread through air spaces (STAS) is defined as the migration of tumor cells into the surrounding lung tissue - specifically the alveolar spaces - beyond the visible edge of the main tumor mass. Recognized by the World Health Organization in 2015, STAS represents an important mode of lung cancer progression.
STAS worsens outcomes significantly. The presence of STAS in surgically removed lung tumors is strongly associated with reduced overall survival and shorter disease-free survival. It is found in nearly 23% of pathologic stage I non-small cell lung cancers, making it a clinically significant finding even in early-stage disease.
A challenge for surgeons. STAS cannot be reliably detected during surgery using frozen section pathology - the rapid tissue analysis performed while a patient is still on the operating table. This limitation means surgeons often cannot adjust their approach in real time based on STAS status, making preoperative prediction especially valuable.
CT scanning as a preoperative tool. Since STAS status affects surgical decisions - particularly whether limited resection or more extensive surgery is appropriate - identifying CT scan features that can predict STAS before surgery would allow better treatment planning and potentially improve patient outcomes.
A retrospective study of 145 patients. This study reviewed CT scans and surgical pathology data from 145 patients who underwent surgery for stage T1 lung adenocarcinoma between 2017 and 2021. All CT scans were performed within two months before surgery, and pathological STAS assessment was performed on the removed tissue specimens.
Qualitative CT features evaluated. Two radiologists independently assessed a comprehensive set of CT characteristics: nodule type (pure ground-glass, part-solid, or solid), margin appearance, presence of lobulation, spiculation, cavity formation, calcification, central low attenuation, air bronchogram, satellite lesions, pleural retraction, and background lung abnormalities.
Quantitative measurements using AI-assisted software. Advanced 3D CT analysis software (SYNAPSE VINCENT) was used to automatically measure tumor volume, solid component volume and percentage, and CT density values (maximum, minimum, and mean in Hounsfield units). Radiologists also measured tumor and solid component diameters in multiple planes.
Statistical analysis approach. CT features were compared between the 64 STAS-positive and 81 STAS-negative patients. ROC curve analysis identified optimal cutoff values, and multiple logistic regression identified the strongest independent predictor of STAS. Multicollinearity testing ensured the statistical validity of the final model.
Central low attenuation: the strongest indicator. The most statistically powerful predictor of STAS was central low attenuation - a dark area within the solid part of the tumor visible on CT. Patients with this finding were nearly four times more likely to have STAS (odds ratio 3.993), and the difference between STAS-positive (64%) and STAS-negative (31%) cases was highly significant.
Lobulation associated with STAS. The characteristic lobulated shape of a tumor's border - reflecting uneven growth or mixed tissue composition - was significantly more common in STAS-positive cases (62.5% vs. 38.3%). This feature reflects the aggressive, irregular growth pattern typical of tumors with airspace spread.
Air bronchogram signals absence of STAS. Interestingly, the presence of air bronchograms - air-filled airways visible within a tumor on CT - was significantly more frequent in STAS-negative cases (34.6% vs. 14.1%), effectively serving as a reassuring sign against STAS. This inverse association highlights how CT patterns encode meaningful biological information.
Larger solid component in STAS-positive tumors. Quantitative analysis showed that the solid component volume (2187 vs. 1116 cubic mm), the percentage of solid tissue (58% vs. 41%), and maximum solid diameter (18 vs. 13 mm) were all significantly greater in STAS-positive cases, even when the overall tumor size was not significantly different.
Solid volume showed highest individual predictive power. Among all individual quantitative measures, solid component volume had the highest area under the ROC curve (AUC of 0.687), with an optimal cutoff of 2055 cubic mm. At this threshold, it achieved 90% specificity - meaning very few patients without STAS would be incorrectly classified as positive.
Combining multiple features improves accuracy. When multiple CT variables were combined - including mean CT value, solid volume, solid percentage, max solid diameter, lobulation, central low attenuation, and absence of air bronchogram - the AUC improved to 0.731. This demonstrates that integrating both qualitative and quantitative information provides better predictive discrimination than any single feature alone.
Preoperative biopsy associated with STAS status. Patients who underwent preoperative biopsy were significantly more likely to be STAS-positive (p=0.041), reflecting that clinicians tend to biopsy tumors with larger solid components. However, biopsy results themselves (positive or negative for cancer) were not independently associated with STAS status.
Lymph node metastasis more common with STAS. Lymph node involvement was significantly more frequent in STAS-positive patients (14.1% vs. 2.5%), underscoring the clinical significance of STAS as a marker of more aggressive disease behavior and early metastatic spread in otherwise early-stage tumors.
Why central low attenuation matters. Central low attenuation within a tumor - the dark core visible on mediastinal window CT settings - is pathologically associated with mucinous features. Mucinous adenocarcinomas are known to exhibit higher rates of airspace spread, which likely explains why this finding is such a strong predictor of STAS.
Lobulation reflects aggressive biology. The lobulated contour of a tumor - the bumpy, irregular border - typically reflects different rates of tissue growth or mixed tissue composition within the same mass. This pattern is more common in poorly differentiated and aggressive adenocarcinomas, explaining its association with STAS in this study.
STAS and surgical planning. Because STAS is a significant risk factor for recurrence after limited resection surgery, and tumors with margins smaller than 1 cm are at particularly high risk, preoperative CT prediction of STAS could guide surgeons to perform more extensive operations when features suggest STAS is present, potentially reducing recurrence rates.
Implications beyond surgery. The study authors note that STAS prediction may also be relevant for other local therapies such as stereotactic body radiotherapy and image-guided ablation, where adequate treatment margins are critical to preventing local tumor recurrence. CT-based STAS prediction could thus help optimize non-surgical treatment planning as well.
Study findings summary. This study demonstrated that both qualitative and quantitative CT analysis can meaningfully predict STAS in early-stage lung adenocarcinoma. Central low attenuation emerged as the single strongest independent predictor, and combining multiple CT features further improved diagnostic accuracy.
Key limitations to consider. The study was retrospective and conducted at a single institution with a relatively small patient sample (145 patients), which limits statistical power and generalizability. The study also only evaluated T1-stage adenocarcinomas, leaving open questions about STAS prediction in more advanced tumors or other histological types.
Need for validation. Prospective, multi-institution studies with larger patient cohorts are needed to validate these CT-based STAS predictors before they can be reliably applied in clinical practice. The authors also suggest that future work should examine advanced tumors beyond the T1 stage to better predict prognosis and recurrence risk across a broader patient population.
Potential for AI integration. The use of AI-driven CT volumetry software in this study represents a growing trend in radiology. Future studies may integrate machine learning algorithms that simultaneously process all relevant CT features to provide automated, reproducible STAS risk estimates, moving toward more efficient and standardized preoperative assessment.