Segmenting brain tumors accurately from MRI scans is a challenging task in medical imaging. It plays an important role in ensuring accurate diagnosis and planning effective treatment. This study investigates and compares the SegResNet and 3D UNet, two advanced deep learning architectures, for brain tumor segmentation using the MONAI framework. The performance of these two architectures is evaluated and compared by using two different optimizers (a) Adam, a well-established optimization algorithm in deep learning, and (b) Ranger21, a newer optimizer designed to enhance training stability and generalization. In this study, we follow a technical workflow that starts with data preprocessing, model architecture design, training procedures, and evaluation metrics such as Dice Score, Intersection over Union (IoU), precision, Recall and F1 Score. Through a comparative analysis of these models and optimizers, we assess their accuracy and potential for reliable segmentation. The results demonstrate the effectiveness of the SegResNet and 3D UNet architectures, yielding high accuracy and robust segmentation performance, comparing their potential for improving clinical outcomes in tumor segmentation. The comparative analysis insights into the strengths of both SegResNet and 3D UNet indicate that these models can achieve high accuracy in tumor segmentation. This promising performance suggests that these models, especially when paired with the right optimizer, hold great potential for improving clinical outcomes by providing more reliable and precise tumor segmentation.