Related Experiment Video
Updated: Jun 5, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
MSR-UNet: enhancing multi-scale and long-range dependencies in medical image segmentation.
Shuai Wang1, Lei Liu1,2, Jun Wang3
1School of Computer Science and Technology, Huaibei Normal University, Huaibei, China.
Peerj. Computer Science
|December 9, 2024
Summary
This study introduces a novel multi-scale reconfiguration self-attention (MSR-SA) module for transformer-based medical image segmentation. The MSR-SA module effectively captures multi-scale and long-range dependencies, improving segmentation accuracy.
Area of Science:
- Medical imaging
- Computer vision
- Artificial intelligence
Background:
- Transformer models are vital for medical image segmentation.
- Effective segmentation requires modeling multi-scale information and long-range pixel dependencies.
- Current methods often use fixed single-scale windows, limiting performance.
Purpose of the Study:
- To propose a novel module for enhanced medical image segmentation.
- To address limitations of fixed single-scale windows in transformer models.
- To improve the modeling of multi-scale and long-range dependencies.
Main Methods:
- Introduced a multi-scale reconfiguration self-attention (MSR-SA) module.
- MSR-SA divides attention heads into groups with ascending dilation rates.
- Utilized dilated sampling and dynamic information fusion for multi-scale and long-range feature extraction.
Main Results:
- Developed the multi-scale reconfiguration U-Net (MSR-UNet) framework.
- Achieved satisfactory segmentation results on Synapse and ACDC datasets.
- Demonstrated improved performance in capturing multi-scale and long-range relationships.
Conclusions:
- The MSR-SA module effectively models multi-scale and long-range dependencies.
- MSR-UNet shows promise for accurate medical image segmentation.
- The proposed method offers a potential advancement in the field.
