DE-UFormer: U-shaped dual encoder architectures for brain tumor segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Brain tumor segmentation is critical for diagnosis and treatment planning.
• Current methods using convolutional neural networks (CNNs) or transformers have limitations in capturing both local and global features.
• High-precision local and global contextual information are essential for accurate brain tumor segmentation.

Purpose of the Study:

• To propose a novel brain tumor segmentation model, DE-Uformer, that simultaneously extracts and fuses high-precision local and global contextual information.
• To enhance the accuracy and reliability of automated brain tumor segmentation.

Main Methods:

• Developed the DE-Uformer network model with dual encoders (CNN and Transformer) for comprehensive feature extraction.
• Introduced the nested encoder-aware feature fusion (NEaFF) module for effective deep fusion of multi-dimensional features.
• Utilized spatial attention Transformer and cross-encoder attention Transformer for capturing dependencies within and between encoders.

Main Results:

• The DE-Uformer model demonstrated significantly superior performance compared to state-of-the-art methods on the BraTS2020 and a private meningioma dataset.
• Achieved substantial improvements in brain tumor segmentation accuracy.

Conclusions:

• The proposed DE-Uformer model effectively integrates local and global features for enhanced brain tumor segmentation.
• This advancement holds significant implications for improving diagnostic accuracy, treatment strategy selection, and preoperative planning in neuro-oncology.