PPNet: Pyramid pooling based network for polyp segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Colonoscopy is crucial for gastrointestinal tract investigation and polyp detection.
• Accurate polyp segmentation is vital for effective colonoscopy screening and subsequent treatments like polyp resection.
• Current deep learning methods for polyp segmentation face challenges in achieving precise localization.

Purpose of the Study:

• To propose a novel deep learning network, PPNet, for enhanced polyp segmentation in colonoscopy images.
• To leverage the strengths of Pyramid Pooling Transformer (P2T) for feature extraction and pyramid pooling for feature enhancement.
• To improve the accuracy and effectiveness of polyp segmentation in colonoscopy procedures.

Main Methods:

• Developed PPNet, a network incorporating P2T as an encoder for robust feature extraction.
• Introduced a Pyramid Feature Fusion Module (PFFM) with channel attention for global contextual feature learning.
• Integrated a Memory-keeping Pyramid Pooling Module (MPPM) to enhance layer-by-layer feature extraction in the decoder.

Main Results:

• PPNet demonstrated superior performance compared to several state-of-the-art polyp segmentation networks.
• Experiments conducted on five public colorectal polyp segmentation datasets validated the model's effectiveness.
• The proposed pyramid pooling mechanism significantly contributed to improved colorectal polyp segmentation.

Conclusions:

• PPNet offers a promising approach for precise polyp segmentation in colonoscopy.
• The integration of pyramid pooling and transformer-based features enhances the model's ability to capture contextual information.
• This work highlights the potential of pyramid pooling mechanisms in advancing colorectal polyp segmentation for better patient outcomes.