WCLC 2025 - Posters & ePosters-P1.07.56 Combining Lymph Nodes and Tumor Radiomics Enhances Immunotherapy Efficacy Prediction in Non-Small Cell Lung Cancer

P1.07.56 Combining Lymph Nodes and Tumor Radiomics Enhances Immunotherapy Efficacy Prediction in Non-Small Cell Lung Cancer

Back to course

Pdf Summary

This study investigates the use of combined radiomics from primary tumors, peritumoral regions, and lymph nodes—especially subcarinal nodes—to predict immunotherapy efficacy in non-small cell lung cancer (NSCLC). Analyzing retrospectively 208 NSCLC patients who received immune checkpoint inhibitors (ICI) at Guangdong Provincial People's Hospital, with external validation on 38 patients from Jiangxi Provincial Cancer Hospital, the research extracted radiomic features using the PyRadiomics software. Five machine learning models with two-stage feature optimization were compared, with logistic regression yielding optimal results.<br /><br />The final combined radiomics model (CRM), integrating features from tumor sites and adjacent lymph nodes, demonstrated the best predictive performance, achieving an area under the curve (AUC) of 0.71 on training data, 0.69 internally, and 0.52 externally. Despite incorporating subcarinal lymph nodes—critical in the immunotherapy context—these contributed only minimal gain to prediction accuracy (ΔAUC 0.01), likely limited by CT imaging artifacts and low heterogeneity in lymph node features.<br /><br />Importantly, histologic subtype influenced response, with lung adenocarcinoma showing superior immunotherapy outcomes, supporting treatment decisions informed by tumor histology. The CRM outperformed models relying on single anatomical regions, showing clinical utility for screening patients for ICI therapy. However, the study highlights the need to refine models further, emphasizing quality of radiomic features over broad anatomical coverage to enhance predictive accuracy.<br /><br />In conclusion, combining tumor and lymph node radiomics improves immunotherapy response prediction in NSCLC but requires further optimization for clinical deployment. Future efforts should focus on improving feature extraction quality and validating models across larger, independent cohorts to better guide personalized immunotherapy decisions.

Asset Subtitle

Weijie Zhan

Meta Tag

Speaker Weijie Zhan

Topic Early-Stage Non-small Cell Lung Cancer

Keywords

non-small cell lung cancer

immunotherapy prediction

radiomics

primary tumor

peritumoral region

lymph nodes

subcarinal nodes

machine learning

logistic regression

immune checkpoint inhibitors