NSCLC EGFR Mutation Prediction via Random Forest Model: A Clinical-CT-Radiomics Integration Approach.

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Non-small cell lung cancer (NSCLC) is a leading cause of cancer mortality globally.
• Accurate epidermal growth factor receptor (EGFR) mutation status is crucial for targeted therapy selection with tyrosine kinase inhibitors (TKIs).
• Invasive tissue genotyping for EGFR status can be limited by accessibility and sample quality.

Purpose of the Study:

• To develop and validate a non-invasive machine learning model for predicting EGFR mutation status in NSCLC patients.
• To integrate clinical data, CT morphological features, and radiomic descriptors for enhanced prediction accuracy.
• To assess the clinical utility of the developed model as a screening tool.

Main Methods:

• A retrospective analysis of 138 NSCLC patients with confirmed EGFR status and pre-treatment CT scans.
• Extraction of radiomic features using PyRadiomics, followed by feature selection via mutual information, Spearman correlation, and wrapper methods.
• Training and evaluation of five Random Forest models, with the best model selected based on performance metrics.

Main Results:

• The optimal machine learning model, utilizing 11 selected variables, achieved an Area Under the Curve (AUC) of 0.91 and an accuracy of 0.88 ± 0.03.
• Subgroup analysis demonstrated high performance for wild-type (WT) EGFR prediction (precision 0.93, recall 0.92, F1-score 0.91) and good performance for mutant EGFR prediction (precision 0.76, recall 0.71, F1-score 0.68).
• SHapley Additive exPlanations (SHAP) identified tobacco use, enhancement pattern, and gray-level-zone entropy as significant predictors; decision curve analysis confirmed clinical utility.

Conclusions:

• A non-invasive machine learning model combining clinical, CT morphological, and radiomic features can accurately predict EGFR mutation status in NSCLC.
• This model shows significant potential as a non-invasive screening tool, potentially reducing the need for invasive procedures.
• The findings support the integration of AI-driven radiomics in personalized NSCLC treatment strategies.