Purpose To investigate the feasibility of multi-sequence MRI-based radiomics for predicting epidermal growth factor receptor(EGFR)mutation status in brain metastases from non-small cell lung cancer(NSCLC).Materials and Methods This retrospective study included 237 patients with NSCLC brain metastases from the Second Xiangya Hospital of Central South University(January 1,2017 to December 31,2023)who underwent EGFR genetic testing.All patients underwent pretreatment brain MRI including contrast-enhanced T1-weighted,T2-weighted FLAIR,and T2-weighted sequences,along with chest CT for primary lung lesions.EGFR mutations were identified in 120 patients.Using December 31,2021 as the cutoff date,patients were divided into training(n=146)and validation(n=91)cohorts.Senior radiologists delineated brain metastases on multi-sequence MRI and primary lesions on CT.A total of 851 radiomic features were extracted using PyRadiomics.Following feature selection,machine learning models were constructed using support vector machine algorithm and compared with least absolute shrinkage and selection operator-derived radiomic signatures.Five models were developed:three single-sequence MRI models,a multi-sequence MRI fusion model,and a CT model,with diagnostic performance evaluated by area under the receiver operating characteristic curve.Results The multi-sequence MRI fusion model demonstrated superior performance across all imaging types.The least absolute shrinkage and selection operator and support vector machine models achieved training set area under the curve of 0.854(95%CI 0.748-0.960)and 0.948(95%CI 0.923-0.973),respectively,and validation set area under the curve of 0.810(95%CI 0.751-0.869)and 0.951(95%CI 0.917-0.985),respectively.The optimal prediction model utilized support vector machine algorithm with multi-sequence MRI features.Conclusion Pretreatment multi-sequence MRI radiomics combined with machine learning accurately predicts EGFR mutation status in NSCLC patients with brain metastases.

Purpose To investigate the feasibility of multi-sequence MRI-based radiomics for predicting epidermal growth factor receptor(EGFR)mutation status in brain metastases from non-small cell lung cancer(NSCLC).Materials and Methods This retrospective study included 237 patients with NSCLC brain metastases from the Second Xiangya Hospital of Central South University(January 1,2017 to December 31,2023)who underwent EGFR genetic testing.All patients underwent pretreatment brain MRI including contrast-enhanced T1-weighted,T2-weighted FLAIR,and T2-weighted sequences,along with chest CT for primary lung lesions.EGFR mutations were identified in 120 patients.Using December 31,2021 as the cutoff date,patients were divided into training(n=146)and validation(n=91)cohorts.Senior radiologists delineated brain metastases on multi-sequence MRI and primary lesions on CT.A total of 851 radiomic features were extracted using PyRadiomics.Following feature selection,machine learning models were constructed using support vector machine algorithm and compared with least absolute shrinkage and selection operator-derived radiomic signatures.Five models were developed:three single-sequence MRI models,a multi-sequence MRI fusion model,and a CT model,with diagnostic performance evaluated by area under the receiver operating characteristic curve.Results The multi-sequence MRI fusion model demonstrated superior performance across all imaging types.The least absolute shrinkage and selection operator and support vector machine models achieved training set area under the curve of 0.854(95%CI 0.748-0.960)and 0.948(95%CI 0.923-0.973),respectively,and validation set area under the curve of 0.810(95%CI 0.751-0.869)and 0.951(95%CI 0.917-0.985),respectively.The optimal prediction model utilized support vector machine algorithm with multi-sequence MRI features.Conclusion Pretreatment multi-sequence MRI radiomics combined with machine learning accurately predicts EGFR mutation status in NSCLC patients with brain metastases.