Related Experiment Video
Updated: Jan 10, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.3K
SEFormer for medical image segmentation with integrated global and local features
Chen Ge1, Haoze Pan2, Yihua Song3,4
1Shandong University of Engineering and Vocational Technology, Jinan, 250200, China.
Scientific Reports
|November 25, 2025
Summary
This study introduces SEFormer, a novel medical image segmentation method. It enhances accuracy by combining Convolutional Neural Networks (CNNs) with Transformers and SE fusion for superior local and global feature representation.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Existing medical image segmentation methods struggle to simultaneously capture local and global features.
- Transformer and CNN-based approaches have limitations in representing both feature types effectively.
Purpose of the Study:
- To propose a novel hybrid network architecture, SEFormer, for enhanced medical image segmentation.
- To improve segmentation accuracy and efficiency by effectively leveraging local and global feature representations.
Main Methods:
- Developed a hybrid network combining SENet, ResNet (CNNs), and Transformer components.
- Utilized SE fusion within a feature pyramid structure to integrate local and global features at each layer.
- Incorporated image pyramid concepts to achieve a larger receptive field and prevent feature loss.
Main Results:
- SEFormer demonstrated superior performance in medical image segmentation tasks.
- Achieved a 3.25% improvement in segmentation accuracy on the CHASEDB dataset compared to existing methods.
- The hybrid architecture effectively captured both local and global features, enhancing segmentation quality.
Conclusions:
- SEFormer offers a significant advancement in medical image segmentation.
- The proposed hybrid approach effectively addresses limitations of prior methods in feature representation.
- This method holds promise for improving diagnostic accuracy through enhanced medical image analysis.
