Somatic Mutations Drive Distinct Imaging Phenotypes in Lung Cancer

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence
• Genomics

Background:

• Somatic mutations drive tumor development and phenotype.
• Radiographic phenotypes are typically disconnected from specific mutations.
• Artificial intelligence, specifically radiomics, quantifies imaging phenotypes.

Purpose of the Study:

• To connect imaging phenotypes with somatic mutations in lung adenocarcinoma.
• To develop and validate radiomic signatures for predicting tumor genotypes.
• To assess the performance of radiomic signatures against conventional predictors.

Main Methods:

• Integrated analysis of independent datasets (763 lung adenocarcinoma patients).
• Somatic mutation testing and engineered CT image analytics.
• Development and validation of radiomic signatures in discovery and validation cohorts.

Main Results:

• Radiomic signatures outperformed conventional radiographic predictors.
• A signature for radiographic heterogeneity distinguished EGFR+ from EGFR- cases (AUC=0.69).
• A signature distinguished EGFR+ from KRAS+ tumors (AUC=0.80), improving prediction when combined with clinical models (AUC=0.86).

Conclusions:

• Somatic mutations drive distinct radiographic phenotypes predictable by radiomics.
• Radiomics can serve as noninvasive, low-cost imaging biomarkers.
• This approach has significant clinical implications for personalized medicine.