The rapid advancement of technologies is transforming brain tumor diagnosis, with emphasis on precise segmentation for early intervention. Traditional manual approaches struggle with noise and intensity variations in medical imaging. Deep learning methodologies, particularly UNet++, have gained prominence. We propose an enhanced framework integrating Swin Transformer as the encoder of UNet++ architecture, incorporating Squeeze-and-Excitation (SE) Blocks to refine feature representation. The Swin Transformer captures global contextual dependencies while SE Blocks enhance channel-wise attention. The UNet++ decoder facilitates multi-scale feature fusion through skip connections. We evaluate our model on the Kaggle lower-grade gliomas dataset (110 patient scans), demonstrating superior performance with a Dice coefficient of 0.9643 and mean IoU of 0.945, outperforming previous methodologies. Our model effectively delineates tumor boundaries, potentially corroborating clinical judgment and influencing patient outcomes.