Related Experiment Video
Updated: Jun 18, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
381
BMCS-Net: A Bi-directional multi-scale cascaded segmentation network based on transformer-guided feature Aggregation
Bicao Li1, Jing Wang1, Bei Wang2
1School of Electronic and Information Engineering, Zhongyuan University of Technology, Zhengzhou, 450007, China.
Computers in Biology and Medicine
|July 30, 2024
Summary
This study introduces BMCS-Net, a novel network combining CNNs and Transformers for medical image segmentation. It effectively captures both local details and global context, improving segmentation accuracy for skin and polyp datasets.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Convolutional Neural Networks (CNNs) are effective for medical image segmentation but struggle with long-range dependencies, often losing localized details due to deep architectures and downsampling.
- Transformers excel at modeling global features and long-range correlations, offering a complementary approach to CNNs.
Purpose of the Study:
- To propose a novel Bi-directional Multi-scale Cascaded Segmentation Network (BMCS-Net) that integrates CNN and Transformer features for enhanced medical image segmentation.
- To address the limitations of CNNs in capturing long-range relations and the loss of localized details.
Main Methods:
- Developed BMCS-Net, a hybrid network leveraging both CNNs and Transformers.
- Introduced a Two-stream Cascaded Feature Aggregation (TCFA) module to effectively fuse features from CNN and Transformer branches, preventing global information dilution.
- Incorporated a Multi-Scale Expansion-Aware (MSEA) module to capture contextual information and preserve fine details.
Main Results:
- BMCS-Net demonstrated excellent performance in medical image segmentation tasks.
- The network achieved superior results on both skin and polyp segmentation datasets.
- The proposed TCFA and MSEA modules effectively addressed the limitations of existing methods.
Conclusions:
- BMCS-Net successfully integrates the strengths of CNNs and Transformers for superior medical image segmentation.
- The novel feature aggregation and context-aware modules significantly improve the model's ability to handle both global and local image information.
- BMCS-Net offers a promising approach for accurate disease diagnosis and clinical research through improved segmentation.
