Prediction model for epidermal growth factor receptor mutation status in lung cancer brain metastases with deep learning methods based on dual-modality MRI

Objectives: The preoperative diagnose of epidermal growth factor receptor (EGFR) is crucial for targeted therapy in non-small cell lung cancer (NSCLC) patients with brain metastases. This study aimed to assess the feasibility of using deep learning method to predict EGFR mutation status based on multi-modal MRI in NSCLC brain metastasis patients. Materials and Methods: This study involved 96 patients with pathologically confirmed NSCLC brain metastases from the main cohort and 15 patients from an external cohort. Feature extraction was performed on T1ce and T2-Flair sequences using pre-trained EfficientNetB2 and VGG16 deep learning networks. A multi-step feature reduction process was applied, and classification models were constructed using mainstream machine learning algorithms. We compared the predictive efficacy across different imaging modalities, conducted deep learning and traditional radiomics analysis. Confusion matrix, receiver operating characteristic curve (ROC), area under the curve (AUC), decision curve analysis (DCA), and the Delong test were used to compare model efficiency. Results: The AUCs for the T1ce DL Model, T2-Flair DL Model, and Hybrid Model on the internal test set were 0.987, 0.985 and 0.971, respectively, significantly outperforming their corresponding radiomics models (P < 0.01). The T1ce DL Model also demonstrated strong generalization ability, with an AUC of 0.778 on the external test set. Conclusion: T1ce-based models exhibited superior predictive performance and showed excellent generalizability. The proposed model has potential clinical application value for accurate prediction of EGFR mutation status in NSCLC brain metastases patients.