Improved Brain Tumor Segmentation

Brain tumors are difficult to segment due to their complex shapes and varying appearances. Accurate segmentation is needed for diagnosis, treatment planning, and follow-up. This study is focused on improving brain tumor segmentation using the Swin U-Net model, which utilizes the Swin Transformer’s ability to learn global features and the U-Net’s powerful seg- mentation capability. MRI scans from benchmark datasets like Brats2024 utilized for testing and training the model. The Swin U-Net is based on an encoder-decoder structure, in which the encoder takes fine details using Swin Transformer blocks and the decoder builds high-resolution segmentation maps. Experiments showed significant improvements in segmentation accuracy, with larger Dice coefficients than standard convolutional neural net- works. The model correctly identified tumor boundaries and generalizes well between tumors and imaging scenarios. These findings show the potential of the Swin U-Net model as a high- performance automated brain tumor segmentation tool for more accurate diagnosis and personalized treatment planning in neuro-oncology.