Related Experiment Video
Updated: Aug 7, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
471
EG-TransUNet: a transformer-based U-Net with enhanced and guided models for biomedical image segmentation
Shaoming Pan1, Xin Liu1, Ningdi Xie1
1The State Key Laboratory of Information Engineering in Surveying, Mapping, and Remote Sensing, Wuhan University, Wuhan, China.
BMC Bioinformatics
|March 7, 2023
Summary
This study introduces EG-TransUNet, an advanced deep learning model for biomedical image segmentation. It effectively addresses challenges in feature extraction and information fusion, improving lesion segmentation accuracy.
Area of Science:
- Biomedical Image Analysis
- Deep Learning
- Computer Vision
Background:
- Deep learning methods, particularly convolutional neural networks (CNNs), have advanced biomedical image segmentation.
- Challenges remain in extracting discriminative features from lesions with variable sizes/shapes and fusing spatial/semantic information due to redundant data and semantic gaps.
Purpose of the Study:
- To propose an improved deep learning architecture for biomedical image segmentation.
- To enhance feature discrimination and information fusion for more accurate lesion identification.
Main Methods:
- Utilized attention-based Transformer in both encoder and decoder stages for improved feature extraction.
- Introduced EG-TransUNet architecture incorporating a progressive enhancement module, channel spatial attention, and semantic guidance attention.
- Employed multi-head self-attention to capture object variabilities.
Main Results:
- EG-TransUNet demonstrated superior performance on various biomedical datasets.
- Achieved high mDice scores of 93.44% on Kvasir-SEG and 95.26% on CVC-ClinicDB colonoscopy datasets.
- Showcased improved generalization ability across five medical segmentation datasets.
Conclusions:
- The proposed EG-TransUNet effectively captures object variabilities in medical images.
- EG-TransUNet advances the state-of-the-art in biomedical image segmentation, offering better accuracy and generalization.
- The attention-based Transformer modules significantly contribute to improved feature discrimination and information fusion.
