The segmentation of 3D brain tumors in medical imaging plays a pivotal role in accurate diagnosis and treatment planning. This study focuses on the development and evaluation of a volumetric segmentation model for 3D brain tumor images using advanced techniques in deep learning and Dynamic Slice Selection Technique. The challenges posed by variations in data and complex tumor shapes are addressed through leveraging the capabilities of a 3D UNet architecture. The objectives include enhancing time efficiency, reducing computational complexity, and achieving effective segmentation. Utilizing a dataset consisting of 498 MR images, the study compares the implemented model with a base paper model. Evaluation metrics such as DICE scores for different tumor regions (whole tumor, enhancing tumor, and tumor core), sensitivity, specificity, IOU, loss, and accuracy are employed. The training process involves the Dynamic Slice Selection technique from the 3D tumor data, and results obtained demonstrate the impact of varying the number of slices on segmentation performance. The findings underscore the significance of employing Dynamic 3DUNet for volumetric segmentation, showcasing improvements in Dice Score for EDEMA—0.87, Enhancing—0.86 and Necrotic—0.83.