Related Experiment Video
Updated: Jul 1, 2025
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
399
Simultaneous segmentation and classification of colon cancer polyp images using a dual branch multi-task learning
Chenqian Li1,2, Jun Liu1,2, Jinshan Tang3
1School of Computer Science and Technology, Wuhan University of Science and Technology, Wuhan 430065, China.
Mathematical Biosciences and Engineering : MBE
|March 8, 2024
Summary
This study introduces a novel multi-task network for simultaneous polyp segmentation and classification, improving colorectal cancer diagnosis. The new model effectively handles polyp variations and enhances boundary details for more accurate results.
Area of Science:
- Medical imaging analysis
- Artificial intelligence in healthcare
- Computational pathology
Background:
- Accurate polyp segmentation and classification are crucial for colorectal cancer diagnosis and treatment.
- Existing methods often address segmentation and classification separately, limiting their effectiveness due to task correlations.
- Polyps present challenges like random regions, varied shapes/sizes, and similar boundaries/backgrounds, which current models struggle to address robustly.
Purpose of the Study:
- To develop a robust multi-task network for simultaneous polyp segmentation and classification.
- To leverage the correlation between segmentation and classification tasks for improved accuracy.
- To effectively handle the inherent complexities of polyp appearance in medical images.
Main Methods:
- A dual-branch network combining a transformer and a convolutional neural network (CNN) for enhanced local and global feature representation.
- A feature interaction module (FIM) to bridge semantic gaps and integrate information between the transformer and CNN branches.
- A reverse attention boundary enhancement (RABE) module to preserve and enhance critical edge details for precise polyp identification.
Main Results:
- The proposed multi-task network demonstrated superior performance in both polyp segmentation and classification tasks.
- Extensive experiments on five public datasets confirmed the method's effectiveness compared to state-of-the-art approaches.
- The dual-branch structure and specialized modules (FIM, RABE) significantly improved the handling of polyp variations and edge details.
Conclusions:
- The developed multi-task network offers a more effective and robust solution for polyp analysis in colorectal cancer screening.
- Simultaneous segmentation and classification, coupled with advanced feature integration and boundary enhancement, lead to improved diagnostic accuracy.
- This approach represents a significant advancement in automated polyp detection and characterization for clinical applications.
