BTSwin-Unet: 3D U-shaped Symmetrical Swin Transformer-based Network for Brain Tumor Segmentation with Self-supervised Pre-training

Medical image automatic segmentation plays an important role in Computer-Aided Diagnosis system. Although convolution-based network has achieved great performance in medical image segmentation, it has limitations in modeling long-range contextual interactions and spatial dependencies. Due to the powerful ability of long-range information interaction of Vision Transformer, Vision Transformer have achieved advanced performance in several downstream tasks via self-supervised learning. In this paper, motivative by Swin Transformer, we proposed BTSwin-Unet, which is a 3D U-shaped symmetrical Swin Transformer-based network for brain tumor segmentation. Moreover, we construct a self-supervised learning framework to pre-train the model encoder through the reconstruction task. Extensive experiments on tumor segmentation tasks validated the performance of our proposed model, and our results consistently demonstrate favorable benchmarks.

Automated summary

This study proposes a BTSwin-Unet network based on Swin Transformer for brain tumor segmentation, with pre-training using a self-supervised learning framework. Extensive experiments validate the model's performance in tumor segmentation tasks, consistently achieving favorable benchmarks.