CT Radiogenomics in Metastatic Lung Adenocarcinoma: Single vs Multi-Site Analysis and Outcomes

Overview

This study applied unsupervised clustering of CT-derived radiomic features from single and multiple metastatic sites in lung adenocarcinoma patients to predict oncogenic alterations and clinical outcomes. Stratification by smoking status revealed distinct imaging-genomic associations, with multi-site radiomics enhancing the prediction of molecular profiles and overall survival.

Background

Lung adenocarcinoma (LUAD) is the most common subtype of lung cancer, often diagnosed at an advanced metastatic stage with poor prognosis. Molecular profiling guides targeted therapies, but repeated biopsies are invasive and resource-intensive. Radiomics, a quantitative imaging technique extracting tumor phenotypes from CT scans, offers a non-invasive biomarker to predict oncogenic alterations. Integrating radiomic data from multiple metastatic sites may improve patient stratification and outcome prediction beyond single-lesion analysis.

Data Highlights

The study included adults with metastatic LUAD having at least two segmentable lesions ≥1 cm³ and routine molecular screening. Radiomic features were extracted from contrast-enhanced CT scans and clustered using unsupervised machine learning. Molecular profiling covered key oncogenic alterations including KRAS, EGFR, ALK, ROS1, and others. Clinical outcomes assessed were overall response rate (ORR) and overall survival (OS).

Key Findings

• Unsupervised clustering of radiomic features identified distinct imaging phenotypes associated with specific oncogenic alterations (OAs) in metastatic LUAD.
• Multi-site radiomics incorporating multiple metastatic lesions improved the prediction accuracy of molecular profiles compared to single-site analysis.
• Stratification by smoking status revealed that non-smoker-related OAs (e.g., EGFR, ALK, ROS1) and smoker-related OAs (e.g., KRAS, BRAF, STK11) corresponded to different radiomic clusters.
• Radiomic clusters correlated with clinical outcomes including treatment response and overall survival, supporting their prognostic value.
• Radiomics provided a non-invasive approach to complement tissue biopsy, potentially reducing the need for repeat invasive sampling in metastatic LUAD.

Clinical Implications

Radiomics-based imaging biomarkers can non-invasively predict oncogenic alterations and stratify metastatic lung adenocarcinoma patients by molecular subtype and prognosis. Incorporating multi-site radiomic data enhances the robustness of these predictions, which may guide personalized treatment decisions and optimize biopsy strategies. This approach supports precision oncology by integrating imaging and molecular data to improve patient management.

Conclusion

CT-based radiogenomics using multi-site radiomic analysis offers a promising tool to predict oncogenic alterations and clinical outcomes in metastatic lung adenocarcinoma. This methodology may facilitate personalized treatment and reduce reliance on invasive biopsies.

References

1. Bergonié Institute, Bordeaux, France -- CT-Based Radiogenomics Evaluation of Metastatic Lung Adenocarcinoma