Boundary Constraint Network With Cross Layer Feature Integration for Polyp Segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate polyp localization in endoscopy is crucial for patient treatment and surgical planning.
• Current deep convolutional neural networks (CNNs) for polyp segmentation have limitations in performance.
• Visual inspection for polyp detection is subjective and labor-intensive.

Purpose of the Study:

• To propose a novel boundary constraint network (BCNet) for accurate polyp segmentation in endoscopy images.
• To improve upon existing CNN-based polyp segmentation methods.
• To enhance the precision of polyp masks for clinical applications.

Main Methods:

• Developed BCNet, a novel boundary constraint network for polyp segmentation.
• Integrated cross-level context information using a Cross-Layer Feature Integration Strategy (CFIS).
• Employed attention-driven modules (ACFIMs, GFIMs) and a bilateral boundary extraction module for enhanced feature fusion and boundary detection.

Main Results:

• BCNet demonstrated superior polyp segmentation performance compared to seven state-of-the-art methods.
• The proposed CFIS effectively fused features from multiple layers.
• The bilateral boundary extraction module improved mask accuracy through joint supervision.

Conclusions:

• BCNet achieves accurate polyp segmentation by leveraging cross-level context and boundary information.
• The novel architecture offers improved effectiveness and generalization capabilities.
• BCNet shows promise for advancing automated polyp analysis in clinical settings.