Brain tumor segmentation is a critical step in the diagnosis and treatment planning of brain tumors, yet manual segmentation remains time-consuming, prone to inter-observer variability, and dependent on clinical expertise. While convolutional neural networks (CNNs) have advanced automated segmentation, their limitations in capturing long-range dependencies and global context hinder performance, especially for irregular and small tumor regions. Addressing these challenges, this study introduces SwinTSDNet, a novel Swin Transformer-based framework designed for accurate brain tumor segmentation. SwinTSDNet leverages hierarchical feature extraction and shifted window self-attention to effectively capture both local and global features, ensuring precise tumor delineation. A hybrid loss function combining Dice loss and cross-entropy loss further optimizes the model for diverse tumor sizes and shapes. The proposed model was evaluated on benchmark datasets and compared with six state-of-the-art transformer-based methods, achieving superior performance with a Dice Similarity Coefficient of 0.910, Intersection over Union of 0.885, and Hausdorff Distance of 3.2. Robustness and sensitivity analyses demonstrated the model’s resilience to noise, intensity variations, and hyperparameter changes. These findings establish SwinTSDNet as a robust and clinically viable solution, offering improved segmentation accuracy, enhanced diagnostic confidence, and potential integration into real-world medical imaging workflows to aid radiologists in early and precise tumor identification.