Related Experiment Video
Updated: May 14, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
DSF-BRNet: Dual-Gated Semantic Fusion and Boundary Refinement for Efficient Endoscopic Polyp Segmentation
Botao Liu1, Changqi Shi1, Ming Zhao2
1School of Computer Science, Yangtze University, Jingzhou 434023, China.
Sensors (Basel, Switzerland)
|May 13, 2026
Summary
A new deep learning model, DSF-BRNet, accurately segments colorectal polyps in colonoscopies. This method improves early cancer detection by enhancing lesion localization and boundary refinement for computer-aided diagnosis.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Accurate colorectal polyp segmentation is vital for colorectal cancer prevention.
- Automated segmentation faces challenges like inter-class variance, complex backgrounds, and blurred boundaries.
Purpose of the Study:
- To develop an efficient and accurate deep learning model for endoscopic polyp segmentation.
- To address limitations in current automated polyp segmentation methods.
Main Methods:
- Introduced Dual-Gated Semantic Fusion (DSF) module for feature alignment and semantic localization.
- Implemented High-Frequency Boundary Refinement (HBR) module for contour sharpening.
- Developed an Align-then-Refine framework for improved segmentation.
Main Results:
- Achieved competitive performance on four public datasets (Kvasir-SEG, CVC-ClinicDB, CVC-ColonDB, ETIS-LaribPolypDB).
- Obtained mean Dice scores of 0.943 on CVC-ClinicDB and 0.818 on ETIS-LaribPolypDB.
- Demonstrated favorable computational efficiency with 25.55 M parameters and 80.08 FPS inference speed.
Conclusions:
- The DSF-BRNet model effectively achieves accurate semantic localization and fine boundary preservation.
- The method shows promise for real-time computer-aided diagnosis (CAD) in colonoscopy.