Deep Learning and Multi-Omics for Lung Cancer TME Risk Stratification

Abstract Objective Immune checkpoint inhibitors are increasingly used in elderly lung cancer patients, yet treatment response varies widely due to tumor immune microenvironment (TME) heterogeneity. This study aimed to develop a CT image-based deep learning model for TME risk stratification in elderly patients (≥ 60 years) and to explore the underlying imaging-transcriptomic mechanisms. Methods Transcriptomic and clinical data for 436 elderly lung cancer patients were obtained from TCGA, with matched CT images from TCIA. Tumor segmentation was performed using nnU-Net, followed by feature extraction via a 3D residual network. The DeepSurv model was constructed combining imaging features with clinical variables. Multi-omics analyses included differential expression, functional enrichment, immune microenvironment assessment (CIBERSORT/ESTIMATE/ssGSEA), WGCNA, and Lasso-Cox regression. External validation used ICGC and GEO immunotherapy datasets. Results The nnU-Net model achieved a Dice coefficient of 0.858. The DeepSurv fusion model attained a C-index of 0.748, outperforming clinical models (0.668). High-risk patients had significantly shorter median overall survival than low-risk patients (13.4 vs. 23.8 months, P < 0.001). A 12-gene signature was identified, enriched in extracellular matrix remodeling, TGF-β signaling, and epithelial-mesenchymal transition. The high-risk group exhibited increased M2 macrophage and regulatory T-cell infiltration with reduced CD8⁺ T cells, indicating an immunosuppressive phenotype. The signature stratified survival in the ICGC cohort (P = 0.002). Imaging-molecular correlation confirmed that imaging features reflect key immunosuppressive mechanisms. Conclusion This CT-based deep learning model effectively identifies elderly lung cancer patients with an immunosuppressive TME, offering superior risk stratification over traditional models. The imaging-molecular correlation supports CT imaging as a potential non-invasive tool for TME assessment and immunotherapy decision-making in elderly patients.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-13
DOI
https://doi.org/10.21203/rs.3.rs-11177318/v1
Primary Topic
Radiomics and Machine Learning in Medical Imaging
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Deep Learning and Multi-Omics for Lung Cancer TME Risk Stratification

Tan Liu, Xiang Lv, Shuanghu Yuan, Kang Wang et al.
4 links
Research Square
Radiomics and Machine Learning in Medical Imaging
preprint

Deep Learning and Multi-Omics for Lung Cancer TME Risk Stratification

Tan Liu, Xiang Lv, Shuanghu Yuan, Kang Wang, Hongping Shi, Hongjing Liu, Xinyue Han
preprint en

Abstract

Abstract Objective Immune checkpoint inhibitors are increasingly used in elderly lung cancer patients, yet treatment response varies widely due to tumor immune microenvironment (TME) heterogeneity. This study aimed to develop a CT image-based deep learning model for TME risk stratification in elderly patients (≥ 60 years) and to explore the underlying imaging-transcriptomic mechanisms. Methods Transcriptomic and clinical data for 436 elderly lung cancer patients were obtained from TCGA, with matched CT images from TCIA. Tumor segmentation was performed using nnU-Net, followed by feature extraction via a 3D residual network. The DeepSurv model was constructed combining imaging features with clinical variables. Multi-omics analyses included differential expression, functional enrichment, immune microenvironment assessment (CIBERSORT/ESTIMATE/ssGSEA), WGCNA, and Lasso-Cox regression. External validation used ICGC and GEO immunotherapy datasets. Results The nnU-Net model achieved a Dice coefficient of 0.858. The DeepSurv fusion model attained a C-index of 0.748, outperforming clinical models (0.668). High-risk patients had significantly shorter median overall survival than low-risk patients (13.4 vs. 23.8 months, P < 0.001). A 12-gene signature was identified, enriched in extracellular matrix remodeling, TGF-β signaling, and epithelial-mesenchymal transition. The high-risk group exhibited increased M2 macrophage and regulatory T-cell infiltration with reduced CD8⁺ T cells, indicating an immunosuppressive phenotype. The signature stratified survival in the ICGC cohort (P = 0.002). Imaging-molecular correlation confirmed that imaging features reflect key immunosuppressive mechanisms. Conclusion This CT-based deep learning model effectively identifies elderly lung cancer patients with an immunosuppressive TME, offering superior risk stratification over traditional models. The imaging-molecular correlation supports CT imaging as a potential non-invasive tool for TME assessment and immunotherapy decision-making in elderly patients.

Research Square
University of Science and Technology of China (CN), Shandong Tumor Hospital (CN), Heze Municipal Hospital (CN)
Good health and well-being
Radiomics and Machine Learning in Medical Imaging
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon