Boundary uncertainty aware network for automated polyp segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Automated colorectal polyp segmentation using deep neural networks is crucial for overcoming limitations of manual visual inspection, such as subjectivity and fatigue.
• Existing methods often struggle with uncertain regions in colonoscopy images, leading to suboptimal segmentation performance.

Purpose of the Study:

• To propose a novel Boundary Uncertainty aware Network (BUNet) for precise and robust colorectal polyp segmentation.
• To enhance the focus on ambiguous areas within colonoscopy images for improved segmentation accuracy.

Main Methods:

• Utilized a pyramid vision transformer encoder for multi-scale feature learning to handle diverse polyp sizes and shapes.
• Introduced a Boundary Exploration Module (BEM) to extract low-level boundary cues.
• Developed a Boundary Uncertainty aware Module (BUM) to focus on error-prone regions using high-level features and boundary information.
• Implemented top-down hybrid deep supervision for coarse-to-fine segmentation.

Main Results:

• The proposed BUNet demonstrated superior performance compared to thirteen existing methods across five public datasets.
• Achieved precise localization of polyp regions by effectively handling boundary uncertainty.
• Showcased strong effectiveness and generalization ability in colorectal polyp segmentation tasks.

Conclusions:

• BUNet offers a significant advancement in automated colorectal polyp segmentation.
• The network's ability to address boundary uncertainty leads to more accurate and reliable polyp detection in colonoscopy.
• The proposed approach holds promise for improving clinical diagnosis and patient outcomes.