Attention based multi-scale parallel network for polyp segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Oncology

Background:

• Colonoscopy is vital for colorectal cancer (CRC) detection and prevention.
• Accurate segmentation of colorectal polyps in images aids clinical diagnosis.
• Challenges include polyp size variation, small polyp size, and resemblance to surrounding tissues.

Purpose of the Study:

• To develop an advanced segmentation method for colorectal polyps.
• To improve the accuracy and reliability of polyp detection in colonoscopy images.
• To address the challenges of segmenting small and visually similar polyps.

Main Methods:

• Proposed a novel parallel network architecture named AMNet (Attention Decoding Multi-scale Network).
• Implemented multi-scale feature fusion using upsampling and downsampling for aggregated high-level information.
• Incorporated a parallel attention module and a reverse fusion module to refine edge details and regional relationships.

Main Results:

• AMNet demonstrated effectiveness in improving polyp segmentation accuracy.
• Extensive experiments were conducted on four public polyp segmentation datasets.
• The proposed method successfully addressed segmentation challenges posed by varying polyp characteristics.

Conclusions:

• AMNet significantly enhances the accuracy of colorectal polyp segmentation.
• The developed network provides valuable assistance for clinicians in diagnosing colorectal conditions.
• This approach contributes to improved colorectal cancer prevention strategies through better polyp identification.