Related Experiment Video
Updated: Jan 16, 2026
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
735
Adjacent-differential network with shallow attention for polyp segmentation in colonoscopy images
Keli Hu1,2,3,4, Chen Wang2, Hancan Zhu2
1Department of Gastroenterology, Cancer Center, Zhejiang Provincial People's Hospital (Affiliated People's Hospital, Hangzhou Medical College), Hangzhou, 310014, People's Republic of China.
Scientific Reports
|September 25, 2025
Summary
Accurate polyp segmentation in colonoscopy images is crucial for early colorectal cancer detection. A new deep learning network, ADSANet, improves polyp segmentation by addressing color inconsistencies and enhancing feature fusion, outperforming existing methods.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Colonoscopy is vital for detecting polyps, which can progress to colorectal cancer.
- Accurate polyp segmentation aids early diagnosis and treatment.
- Existing segmentation methods struggle with inconsistent image colors and suboptimal feature fusion.
Purpose of the Study:
- To develop an advanced deep learning model for accurate polyp segmentation in colonoscopy images.
- To address challenges of color inconsistency and feature fusion in current methods.
Main Methods:
- Proposed ADSANet (deep adjacent-differential network with shallow attention).
- Introduced a color exchange strategy to decouple image content from color variations.
- Developed an adjacent-differential feature fusion module (ADFM) and a shallow attention module (SAM) for enhanced feature synergy.
Main Results:
- ADSANet significantly outperformed state-of-the-art CNN-based methods on five datasets.
- Achieved substantial performance gains over PraNet, ranging from 1.7% to 18.5% across datasets.
- Demonstrated the effectiveness of the proposed color exchange and adjacent-differential feature fusion techniques.
Conclusions:
- ADSANet offers a more accurate polyp segmentation solution for colonoscopy.
- The novel color exchange and feature fusion strategies are key to improved performance.
- This work contributes to advancing early colorectal cancer diagnosis through improved image analysis.
