MRI-based Radiomics signature for the Prediction of Response of Lung Cancer Brain Metastases After Whole-Brain Radiotherapy Background: Local response prediction for brain metastases (BMs) from lung cancer after Whole-Brain Radiotherapy (WBRT) is challenging, as existing criteria are based solely on unidimensional measurements. This study sought to determine whether radiomic features of lung cancer BMs derived from pre-treatment magnetic resonance imaging (MRI) could be used to predict local response following WBRT. Methods: A total of 88 Lung Cancer patients with BMs treated with WBRT were analyzed. After volumes of interest were drawn, 944 radiomic features including first-order, shape, Gray Level Co-occurrence Matrix (GLCM), Gray Level Dependence Matrix (GLDM), Gray Level Run Length Matrix (GLRLM), Gray Level Size Zone Matrix (GLSZM), Neighborhood Gray Tone Difference Matrix (NGTDM), and Laplacian of Gaussian (LoG) features were extracted, using the baseline pre-treatment post-contrast T1 (T1c) and T2 fluid-attenuated inversion recovery (FLAIR) MRI sequences, respectively. Local response status was determined by contrasting the baseline and follow-up MRI according to the RANO-BM criteria. The independent samples t test or Mann-Whitney U test, and then least absolute shrinkage and selection operator (LASSO) were used for dimensionality reduction and feature selection. An adaboost classifier was trained using the selected radiomic features and evaluated using the area under the receiver operating characteristic curve (AUC) in both the training and testing sets. Other discrimination metrics, including classification accuracy, positive predictive value (PPV), negative predictive value (NPV), sensitivity, and specificity, were also calculated. Results: The optimal radiomics signature was developed based on a multivariable logistic regression with 4, 5, 6 radiomic features on T1c, T2 FLAIR and T1c+T2 FLAIR, respectively. The radiomics model based on T1c features presented the AUC of (0.920 vs. 0.805, respectively) for both the training and testing sets, followed by T2 FLAIR features (0.893 vs. 0.701, respectively), and T1c+T2 FLAIR features (0.971 vs. 0.857, respectively). The classification accuracy of the radiomics model also well predicted the local response of BMs for both the the training and testing sets (T1c: 82.9% vs. 77.8%, T2 FLAIR: 82.9% vs. 77.8%, T1c+T2 FLAIR: 90.0% vs. 77.8%, respectively). Conclusions: Radiomics holds promise for predicting local tumor response following WBRT in patients with lung cancer and brain metastases. A predictive model built on radiomic features from an institutional cohort performed well on cross-validation testing. These results warrant further validation in independent datasets. Such work could prove invaluable for guiding management of individual patients and assessing outcomes of novel interventions.