The rapid growth of urban populations demands the integration of advanced technologies to modernize healthcare delivery within smart cities. Telemedicine has become a crucial tool, improving access to and accelerating the delivery of healthcare services. Within this context, medical image segmentation assumes a critical role in enabling precise diagnostic outcomes and facilitating personalized treatment approaches. This study presents a more effective architecture for segmenting brain tumors, built upon a 3D U-Net model enhanced with integrated Squeeze-and-Excitation (SE) blocks. These SE blocks act like attention mechanisms, helping the network focus on the most important features in the images, thus improving segmentation accuracy. Trained using the Medical Segmentation Decathlon (MSD) dataset and employing a robust data augmentation strategy to improve generalization, the proposed model demonstrates significantly improved accuracy in identifying tumor shapes and characteristics. Specifically, it achieves a Dice Similarity Coefficient of 0.996 for edema segmentation, a 46% improvement compared to methods presented in the Medical Segmentation Decathlon (MSD) Challenge. This work advances telemedicine through deep learning-based image segmentation, enabling efficient and scalable smart city healthcare.