Study
NLST
(Learn more about this study)
Project ID
NLST-1470
Initial CDAS Request Approval
Jul 22, 2025
Title
Image-based non-invasive pollutant detection for personalized lung cancer early detection
Summary
Lung adenocarcinoma remains the leading cause of cancer-related deaths, largely due to late diagnosis and suboptimal response to therapies. While CT screening is increasingly adopted for high-risk populations (e.g., smokers), its broader utility is constrained by diagnostic limitations, radiation exposure, and cost. Enhancing clinically relevant interpretation of CT scans is essential to improve cost-effectiveness and the return-on-investment in lung cancer early detection and surveillance. A promising opportunity is to extract individualized pollutant burden from CT. Environmental exposure to airborne pollutants is recognized as a major risk factor for lung cancer, but its individual-level impact is difficult to quantify. Our prior work developed AI tools to measure pollutant burden in lung/tumor tissues and introduced the LPI, which is correlated with inflammation and disease-free survival in lung adenocarcinoma. However, these pathology-based methods require invasive sampling, limiting its application at the population level or in screening settings. Currently, no non-invasive methods exist to quantify individual pollutant exposure. To address this unmet need, we propose a novel image-based “pollutant score”—a non-invasive biomarker that quantifies pulmonary pollutant exposure from standard chest CTs. This approach provides actionable environmental insight in routine screening without added cost, enabling personalized risk assessment and improving the utility of lung cancer screening.
Aims
Managing CT-detected lung nodules remains challenging due to high false-positive rates and ambiguous interpretation. Smoking status alone cannot reliably predict lung cancer risk or outcome, and current screening criteria often overlook the environmental exposure and family history due to complexities. We aim to predict pollutant index (CT-LPI) for each nodule, leveraging the full richness of CT signal to infer environmental context and biological aggressiveness. By integrating CT-LPI with clinical characteristics (e.g., smoking status, family history), we will generate personalized risk scores to guide follow-up recommendations. Our approach bridges histology and radiology to enable non-invasive pollutant quantification from routine imaging. As proof of concept, we analyzed data from an MD Anderson clinical trial, extracting LPI from surgical pathology and radiomic features from pre-surgical CTs. Preliminary results show that lung tumors with higher pollution burden exhibit significantly more compact shapes and reduced texture/density heterogeneity. These findings support the feasibility of using images to infer pollutant-related biological signals, transforming routine CT screening from a detection tool into a personalized stratification platform.
Collaborators
Chengyue Wu, The University of Texas MD Anderson Cancer Center.
Yinyin Yuan, The University of Texas MD Anderson Cancer Center.
Xiaoxi Pan, The University of Texas MD Anderson Cancer Center.
Yutong Li, The University of Texas MD Anderson Cancer Center.
