Brain tumors pose significant challenges in diagnosis and treatment, with precise tumor segmentation being a critical step toward effective management. In this study, we propose a novel approach that combines a customized U-Net (ResNet-152)-based feature extractor with Principal Component Analysis (PCA) for dimensionality reduction and an Artificial Neural Network (ANN) for classification of brain tumors into four categories: Glioma, Meningioma, Pituitary, and No Tumor. The customized U-Net (ResNet-152) architecture enhances the extraction of salient features from Magnetic Resonance Imaging (MRI) scans, while PCA reduces the feature space, optimizing computational efficiency. The ANN classifier then uses these reduced features for accurate tumor classification. To further interpret the model’s decisions, we integrate Explainable AI (XAI) methods, including Grad-CAM for visualizing key regions of interest in MRI scans. Our system achieved an accuracy of 95%, supported by metrics such as precision, recall, and confusion matrices, indicating its robustness and reliability. The proposed model demonstrates significant improvements in classification tasks compared to traditional models, offering a valuable tool for clinical diagnosis and personalized treatment planning. This research highlights the potential of integrating customized U-Net (ResNet-152), PCA, and ANN in advancing precision medicine and supporting clinicians in making informed decisions for brain tumor patients.