A 3D Swin Transformer Fusion Network for Brain Tumor Segmentation

Glioma is a deadly brain tumor, and its diagnosis and treatment are crucial for patients. In this study, we propose a novel automatic brain tumor segmentation model 3D-SwinFusionNet, which combines the features of U-Net and Swin Transformer.By incorporating the Advanced Swin Transformer (AST) and Double Residual Block (DRB), the model improves feature extraction and the capture of local information. Additionally, it includes a segmentation head designed to optimize feature information and reduce decoding errors, thereby enhancing segmentation accuracy to some extent. The experimental results show that our proposed method has the lowest HD value of 2.32 and the highest DICE coefficient of 95.45% The design of 3D-SwinFusionNet performs well in addressing the challenges of medical image segmentation and may help improve the efficiency and accuracy of glioma diagnosis.