An Integrated Clinical-Radiomics-Deep Learning Model Based on ¹⁸F-FDG PET/CT for Predicting EGFR Mutation Status in...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Lung adenocarcinoma (LUAD) treatment is guided by EGFR mutation status.
• Noninvasive prediction of EGFR mutation status is crucial for personalized therapy.

Purpose of the Study:

• To develop and validate an integrated model for predicting EGFR mutation status in LUAD using pretreatment 18F-FDG PET/CT imaging.
• To compare the predictive performance of clinical, clinical-radiomics, and clinical-radiomics-deep learning models.

Main Methods:

• A retrospective analysis of 218 LUAD patients' data, including PET/CT images, clinical characteristics, and EGFR mutation status.
• Development of three models: clinical (C), clinical-radiomics (CR), and clinical-radiomics-deep learning (CRD).
• The CRD model integrated clinical features, ConvNext-based deep learning scores, and LASSO-selected radiomic features.

Main Results:

• The CRD model achieved a superior area under the curve (AUC) of 0.821, outperforming the C model (AUC=0.599) and CR model (AUC=0.739).
• Statistical analysis (DeLong test) confirmed the significant superiority of the CRD model (p<0.001).
• Calibration curves and decision curve analysis demonstrated the CRD model's robustness and clinical utility, leading to a nomogram for individualized risk prediction.

Conclusions:

• Integrating clinical, radiomic, and deep learning features offers a promising noninvasive method for predicting EGFR mutation status in LUAD.
• The developed CRD model shows potential for improving treatment decisions in lung adenocarcinoma patients.