Related Experiment Video
Updated: Aug 4, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
470
MISSFormer: An Effective Transformer for 2D Medical Image Segmentation
IEEE Transactions on Medical Imaging
|April 4, 2023
Summary
MISSFormer, a novel Medical Image Segmentation tranSFormer, enhances feature discrimination by integrating local context and global dependencies. This transformer model achieves superior performance in medical image segmentation tasks.
Area of Science:
- Computer Vision
- Medical Image Analysis
- Artificial Intelligence
Background:
- Transformer models excel at global dependencies in vision tasks.
- Limitations exist in modeling local context and multi-scale feature correlations.
- Existing methods may not fully capture intricate medical image details.
Purpose of the Study:
- Introduce MISSFormer, a hierarchical encoder-decoder transformer for medical image segmentation.
- Address limitations of pure global dependency modeling in transformers.
- Improve feature discrimination and context modeling for enhanced segmentation accuracy.
Main Methods:
- Developed MISSFormer, a hierarchical encoder-decoder network.
- Introduced ReMix-FFN to integrate local context and global dependencies within transformer blocks.
- Proposed a ReMixed Transformer Context Bridge for multi-scale feature correlation extraction.
Main Results:
- MISSFormer demonstrates robust capacity for discriminative feature and context capture.
- Achieved superior performance in multi-organ, cardiac, and retinal vessel segmentation.
- Outperformed state-of-the-art methods, even when trained from scratch.
Conclusions:
- MISSFormer effectively models both local and global information for medical image segmentation.
- The proposed architecture shows significant improvements in segmentation accuracy and robustness.
- The core designs are generalizable to other visual segmentation tasks.
