We present an automated deep learning pipeline for brain MRI analysis that integrates SwinMR-based MRI reconstruction, a hybrid CNN–Transformer segmentation network, and a deep learning classifier for tumor type identification. The framework is evaluated in the BraTS17 and Calgary-Campinas datasets, generating a Dice coefficient of 91. 8% for tumor segmentation and a classification precision of 95. 3%, accurately distinguishing glioma, meningioma, and pituitary tumors. Comparative analysis shows superior performance over baseline U-Net and standard CNN classifiers, and qualitative results highlight improved tumor-boundary delineation and subtype discrimination. These findings validate the effectiveness of combining advanced reconstruction and hybrid deep learning models for robust and reliable brain tumor analysis.