Cascade Aggregation Network for Accurate Polyp Segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate polyp segmentation is vital for computer-aided diagnosis and early colorectal cancer detection.
• Existing Feature Pyramid Networks (FPNs) struggle with detail degradation from upsampling and inadequate global context capture.
• These limitations hinder small polyp segmentation and performance on complex structures.

Purpose of the Study:

• To propose a novel Cascaded Aggregation Network (CANet) for refined polyp segmentation.
• To address the limitations of FPNs in preserving fine details and capturing global context.
• To improve the accuracy of polyp segmentation for enhanced colorectal cancer diagnosis.

Main Methods:

• Utilized a PVT transformer backbone for robust multi-level feature extraction.
• Introduced a Cascade Aggregation Module (CAM) for semantic enrichment without spatial detail loss.
• Integrated a Multiscale Context-Aware Module (MCAM) and Residual-based Fusion Module (RFM) for enhanced feature fusion and context understanding.

Main Results:

• CANet demonstrated superior performance compared to state-of-the-art methods.
• The proposed network effectively preserves spatial details while enriching semantic representations.
• Experiments confirmed CANet's effectiveness in both in-distribution and out-of-distribution scenarios.

Conclusions:

• CANet offers a significant advancement in polyp segmentation technology.
• The network's architecture effectively tackles the inherent limitations of FPNs.
• CANet holds promise for improving the accuracy and reliability of computer-aided diagnosis systems for colorectal cancer.