Related Experiment Video
Updated: Apr 22, 2026
04:48
Swin-PSAxialNet: An Efficient Multi-Organ Segmentation Technique
Published on: July 5, 2024
986
DSCANet: Integrating dual encoder and spatial cross-attention for polyp segmentation
Jun Su1, Tiantian Shi1, Bogdan Adamyk2
1School of Computer Science, Hubei University of Technology, Wuhan, China.
Plos One
|April 20, 2026
Summary
This study introduces DSCANet, a novel dual-branch network for precise colorectal polyp segmentation in colonoscopies. DSCANet effectively fuses body and edge features, improving early cancer detection and diagnostic accuracy.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Gastroenterology
Background:
- Colorectal polyps detected during colonoscopy are precursors to colon cancer, necessitating accurate segmentation for early diagnosis and prevention.
- Current polyp segmentation methods face challenges due to variations in polyp size, shape, and indistinct boundaries, impacting diagnostic reliability.
- Automated and precise polyp segmentation is crucial for enhancing the clinical utility of colonoscopy in cancer screening.
Purpose of the Study:
- To propose a novel deep learning network, DSCANet, for high-precision colorectal polyp segmentation.
- To address the limitations of existing methods in segmenting polyps with diverse characteristics and unclear boundaries.
- To improve the accuracy and efficiency of polyp detection in colonoscopic images.
Main Methods:
- Developed DSCANet, a dual-branch encoder-structured network integrating spatial cross-attention (SCA), bipolar fusion (BF), and flexible axis-attention (FAA) modules.
- The network employs separate encoders for body and edge feature extraction, fused via SCA and BF modules.
- FAA module aids the decoder in leveraging high-level semantic information for enhanced segmentation.
Main Results:
- DSCANet demonstrated superior performance in segmenting colorectal polyps across multiple benchmark datasets.
- The proposed network effectively fused body and edge features, leading to more accurate polyp boundary delineation.
- Achieved high precision in segmenting polyps of varying sizes, shapes, and with poorly defined boundaries.
Conclusions:
- DSCANet offers a robust and effective solution for automated colorectal polyp segmentation.
- The network's architecture facilitates precise feature fusion, significantly improving segmentation accuracy in challenging medical images.
- This advancement holds potential for enhancing colonoscopy's role in colorectal cancer prevention and early diagnosis.
