Related Experiment Video
Updated: May 12, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
320
Hybrid transformer and convolution iteratively optimized pyramid network for brain large deformation image
Xinxin Cui1, Yuee Zhou1, Caihong Wei2
1School of Medical Information Engineering, Gansu University of Traditional Chinese Medicine, Lanzhou, Gansu, 730000, China.
Scientific Reports
|May 5, 2025
Summary
This study introduces a novel hybrid Transformer and convolution network for accurate large deformation brain image registration. The method improves upon existing models by optimizing encoders and decoder features for better multi-scale deformation prediction.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neuroscience
Background:
- Pyramid-based encoder-decoder networks are popular for large deformation image registration.
- Existing methods often overlook encoder impact and lack scale-specific feature map design.
Purpose of the Study:
- To propose an innovative hybrid Transformer and convolution iteratively optimized pyramid network for large deformation brain image registration.
- To address limitations in current registration models by optimizing feature encoders and decoder designs.
Main Methods:
- Designed four encoder variants to assess their impact on registration performance.
- Integrated Swin-Transformer modules with a convolution iterative strategy.
- Tailored each decoder layer based on semantic information characteristics at different scales.
Main Results:
- Achieved the highest registration accuracy across three public brain MRI datasets.
- Outperformed 9 state-of-the-art registration methods in extensive experiments.
Conclusions:
- The proposed hybrid network design demonstrates significant effectiveness for large deformation brain image registration.
- The model shows strong potential for clinical applications in neuroimaging.
