Related Experiment Video
Updated: Jul 1, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
399
Enhancing medical image segmentation with a multi-transformer U-Net
Yongping Dan1, Weishou Jin1, Xuebin Yue2
1School of Electronic and Information, Zhongyuan University Of Technology, Zhengzhou, Henan, China.
Peerj
|March 4, 2024
Summary
This study introduces a hybrid Swin and Deformable Transformer model for enhanced medical image segmentation. The novel approach improves accuracy and speeds up training for CT and X-ray lung images.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Swin Transformer networks show potential in medical image segmentation.
- Existing models face challenges with accuracy and training convergence.
Purpose of the Study:
- To develop a novel hybrid Transformer model for improved medical image segmentation.
- To enhance accuracy and accelerate training convergence in lung image segmentation.
Main Methods:
- Combined Swin Transformer for local features and Deformable Transformer for dynamic sampling.
- Incorporated additional skip connections to minimize information loss.
- Applied the model to CT and X-ray lung image segmentation tasks.
Main Results:
- The hybrid model surpassed standalone Swin Transformer (Swin Unet) performance.
- Achieved accuracy improvements of 0.7% (88.18%) on COVID-19 CT dataset.
- Achieved accuracy improvements of 2.7% (98.01%) on Chest X-ray dataset.
- Demonstrated faster convergence under identical conditions.
Conclusions:
- The proposed hybrid Transformer model offers rapid and accurate segmentation of lung images.
- This advancement can support medical practitioners in early diagnosis and treatment planning.
