Enhancing Brain Tumor Segmentation with Transformer-Based Models: A Study on the BraTS 2020 Dataset

Abstract Accurate segmentation of brain tumors from medical images is crucial for clin- ical diagnosis, treatment planning, and patient outcome prediction. While the UNet architecture has long been a standard in biomedical image segmentation, recent advancements in transformer-based models have opened new possibilities for improving segmentation performance. This study conducts a comparative analysis of three models for brain tumor segmentation: Standard 3D UNet, Swin Transformer-Enhanced 3D UNet, and Detection Transformer-Enhanced 3D UNet. The BraTS 2020 dataset is utilized to train and evaluate these mod- els. Experimental results reveal that the Detection Transformer-Enhanced UNet achieves the highest segmentation accuracy, showcasing its ability to effectively capture long-range dependencies and complex features, surpassing both the stan- dard UNet and the Swin Transformer-Enhanced UNet. These findings highlight the potential of integrating transformer architectures to advance the state of medical image segmentation.