Related Experiment Video
Updated: Aug 2, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
470
High-Resolution Swin Transformer for Automatic Medical Image Segmentation
Chen Wei1, Shenghan Ren2, Kaitai Guo3
1College of Economics and Management, Xi'an University of Posts & Telecommunications, Xi'an 710061, China.
Sensors (Basel, Switzerland)
|April 13, 2023
Summary
This study introduces the high-resolution Swin Transformer network (HRSTNet) for medical image segmentation. HRSTNet maintains high spatial precision, achieving performance comparable to state-of-the-art methods.
Area of Science:
- Medical Image Analysis
- Deep Learning
- Computer Vision
Background:
- Accurate medical image segmentation relies on feature map resolution.
- Existing Transformer-based U-Net architectures may lose spatial precision during high-resolution recovery.
- High-resolution network (HRNet) design principles offer an alternative approach.
Purpose of the Study:
- To develop a Transformer-based network that preserves spatial precision in medical image segmentation.
- To leverage the HRNet design by integrating Transformer blocks for continuous multi-resolution feature exchange.
Main Methods:
- Proposed the high-resolution Swin Transformer network (HRSTNet).
- Replaced convolutional layers with Transformer blocks within an HRNet-style architecture.
- Enabled continuous feature map information exchange across different resolutions.
Main Results:
- HRSTNet demonstrated performance comparable to state-of-the-art Transformer-based U-Net-like architectures.
- Evaluated on diverse datasets: 2021 Brain Tumor Segmentation, Medical Segmentation Decathlon (liver), and BTCV multi-organ segmentation.
Conclusions:
- The HRSTNet effectively addresses the spatial precision limitations of traditional U-Net-like Transformer networks.
- HRSTNet offers a promising alternative for high-accuracy medical image segmentation tasks.
