SwinBTS: A Method for 3D Multimodal Brain Tumor Segmentation Using Swin Transformer.

Area of Science:

• Medical Image Analysis
• Artificial Intelligence in Medicine
• Neuro-oncology Imaging

Background:

• Brain tumor semantic segmentation is crucial for diagnosis and treatment planning.
• Convolutional Neural Networks (CNNs) excel in computer vision but struggle with global context in 3D medical images.
• Transformers offer global information modeling but require integration with CNNs for medical imaging tasks.

Purpose of the Study:

• To develop an advanced 3D brain tumor segmentation method.
• To leverage the strengths of both transformers and CNNs for improved segmentation accuracy.
• To address the limitations of existing methods in capturing both local and global features.

Main Methods:

• Proposed SwinBTS, a hybrid model combining 3D Swin Transformer and CNNs.
• Utilized an encoder-decoder structure with the Swin Transformer for context extraction.
• Employed convolutional operations for efficient downsampling and upsampling.
• Integrated an improved Transformer module for detailed feature extraction.

Main Results:

• SwinBTS demonstrated superior performance in 3D brain tumor segmentation.
• Achieved state-of-the-art results on BraTS 2019, BraTS 2020, and BraTS 2021 datasets.
• Outperformed existing 3D algorithms for brain tumor segmentation on MRI scans.

Conclusions:

• SwinBTS effectively integrates transformer and CNN architectures for 3D medical image segmentation.
• The proposed method significantly improves the accuracy of brain tumor segmentation.
• SwinBTS represents a promising advancement in automated analysis of brain tumor imaging.