Related Experiment Video
Updated: Sep 13, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
523
MFFBi-Unet: Merging Dynamic Sparse Attention and Multi-scale Feature Fusion for Medical Image Segmentation
Baoshan Sun1,2, Chunfei Liu3,4, Qiuyan Wang3,4
1School of Computer Science and Technology, Tiangong University, Tianjin, 300387, China. sunbaoshan@tiangong.edu.cn.
Interdisciplinary Sciences, Computational Life Sciences
|July 29, 2025
Summary
We introduce MFFBi-Unet, a novel deep learning model for medical image segmentation. This architecture uses dynamic sparse attention to improve efficiency and accuracy, outperforming existing methods.
Area of Science:
- Deep Learning
- Medical Image Analysis
- Computer Vision
Background:
- Transformer-based U-Net architectures are effective for medical image segmentation.
- Attention mechanisms in these models cause high computational and memory costs.
- Current sparse attention methods struggle with long-range dependencies.
Purpose of the Study:
- To develop an efficient and accurate medical image segmentation model.
- To address the computational and memory limitations of existing Transformer-based U-Nets.
- To improve the modeling of long-range dependencies in medical images.
Main Methods:
- Proposed MFFBi-Unet architecture with dynamic sparse attention via bi-level routing.
- Integrated BiFormer for optimized semantic feature extraction and reconstruction.
- Introduced a Multi-scale Feature Fusion (MFF) module in skip connections.
Main Results:
- MFFBi-Unet demonstrated significant advantages across multiple public medical benchmarks.
- Achieved statistically significant improvements over state-of-the-art methods.
- Outperformed MISSFormer by 2.02% and 1.28% in Dice scores on respective benchmarks.
Conclusions:
- MFFBi-Unet offers an efficient and adaptable solution for medical image segmentation.
- Dynamic sparse attention and multi-scale feature fusion enhance performance.
- The proposed architecture represents a significant advancement in the field.
