PFPRNet: A Phase-Wise Feature Pyramid With Retention Network for Polyp Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Early detection of colonic polyps is vital for colorectal cancer prevention.
• Deep learning excels at polyp segmentation but faces challenges with limited labeled data and image artifacts like wrinkles.
• Existing methods struggle with integrating global context and local details effectively.

Purpose of the Study:

• To develop an advanced deep learning model for accurate colonic polyp segmentation.
• To address data acquisition limitations and improve model robustness against image complexities.
• To enhance the integration of multi-scale features for better polyp identification.

Main Methods:

• Proposed Phase-wise Feature Pyramid with Retention Network (PFPRNet) utilizing a Transformer-based Encoder.
• Implemented a Phase-wise Feature Pyramid with Retention Decoder for gradual global-local feature integration.
• Introduced an Enhance Perception module and a Low-layer Retention module for improved feature capture and efficient global attention.

Main Results:

• PFPRNet demonstrated strong learning ability and generalization capabilities across multiple polyp segmentation datasets.
• The proposed method outperformed existing state-of-the-art approaches in polyp segmentation accuracy.
• Effective integration of global and local features led to improved attention towards critical polyp regions.

Conclusions:

• PFPRNet offers a promising solution for accurate and efficient colonic polyp segmentation.
• The novel architectural components enhance the model's ability to handle complex polyp image characteristics.
• This approach contributes to advancing early detection strategies for colorectal cancer through improved AI-driven analysis.