DHAFormer: Dual-channel hybrid attention network with transformer for polyp segmentation

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Computational Pathology

Background:

• Accurate early diagnosis of colorectal cancer is crucial.
• Medical image segmentation for polyps is vital for diagnosis.
• Existing segmentation methods face challenges with polyp size, shape, and contrast.

Purpose of the Study:

• To introduce an innovative network for improved polyp segmentation.
• To address limitations of current segmentation techniques in medical imaging.
• To enhance the accuracy of colorectal cancer early diagnosis through better polyp detection.

Main Methods:

• Proposed a Dual-Channel Hybrid Attention Network with Transformer (DHAFormer).
• Incorporated a multi-scale channel fusion module for varied polyp recognition.
• Developed a dual-channel hybrid attention mechanism for enhanced feature representation.

Main Results:

• The DHAFormer framework demonstrated significant improvements in polyp segmentation.
• The multi-scale module effectively recognized polyps of diverse sizes and shapes.
• The attention mechanism successfully reduced background interference and improved foreground polyp features.

Conclusions:

• DHAFormer offers a significant advancement in polyp segmentation accuracy.
• The proposed network effectively tackles challenges of size variation and low contrast in medical images.
• This approach holds promise for improving early colorectal cancer diagnosis.