Radiogenomic Models Using Machine Learning Techniques to Predict EGFR Mutations in Non-Small Cell Lung Cancer

Area of Science:

• Radiomics and Medical Imaging
• Oncology
• Genomics

Background:

• Non-small cell lung cancer (NSCLC) is a leading cause of cancer mortality.
• Epidermal growth factor receptor (EGFR) mutations are key drivers in NSCLC, influencing treatment decisions.
• Accurate identification of EGFR mutation status is crucial for effective precision therapy.

Purpose of the Study:

• To develop radiogenomic models using texture features from CT and 18F-FDG PET-CT images.
• To predict the presence of EGFR mutations in NSCLC patients.
• To differentiate between specific EGFR mutations (exon 19 vs. exon 21).

Main Methods:

• Retrospective analysis of 50 NSCLC patients with known EGFR mutation status.
• Extraction of texture features from pretreatment CT and FDG PET-CT images.
• Development of multivariate logistic regression models to predict EGFR mutations.

Main Results:

• FDG PET-CT texture features achieved an AUC of 0.87 for differentiating EGFR mutant vs. wild type.
• CT texture features achieved an AUC of 0.83 for differentiating EGFR mutant vs. wild type.
• FDG PET-CT texture features showed an AUC of 0.86 for discriminating between EGFR exon 19 and 21 mutations.

Conclusions:

• Texture analysis of CT and FDG PET-CT images can identify EGFR mutations in NSCLC.
• Radiomic signatures show potential for pretreatment assessment and prognosis in precision oncology.
• Imaging-based prediction of EGFR mutations may guide personalized treatment strategies.