Medical imaging, particularly Magnetic Resonance Imaging (MRI), has transformed healthcare by providing high-resolution, detailed visualizations of anatomical structures. MRI generates 3D volumetric data, offering in-depth insights into internal organs, tissues, and abnormalities. In medical image analysis, segmentation is a critical process for accurately delineating anatomical structures or pathological regions within 3D MRI scans. This process is essential for various healthcare applications, including disease monitoring, treatment planning, and clinical diagnosis. This study examines the potential of advanced deep learning architectures for 3D MRI image segmentation. We conduct a comprehensive evaluation of models such as U-Net, HRNet, SegNet, U-Net Residual, and Panoptic Segmentation, assessing their effectiveness in enhancing medical imaging technology. Among these models, U-Net Residual demonstrated outstanding performance, achieving a validation loss of 0.0114, accuracy of 99.77%, Intersection over Union (IoU) of 0.8655, and a Dice coefficient of 0.8362. The Panoptic Segmentation model also exhibited a strong performance, with a validation loss of 0.0083, accuracy of 99.76%, and an IoU score of 0.8476. These findings offer valuable insights for medical imaging practitioners, assisting in the selection of optimal segmentation models for specific applications, ultimately improving diagnostic precision and patient care. Future research could focus on further refining these models and exploring novel methodologies to unlock even greater potential in medical image segmentation.