MEFA-Net: A mask enhanced feature aggregation network for polyp segmentation.

Area of Science:

• Medical imaging
• Computer vision
• Gastroenterology

Background:

• Accurate polyp segmentation is vital for early colorectal cancer diagnosis and treatment.
• Existing methods face challenges like model overfitting, weak generalization, inter-class ambiguity, and intraclass inconsistency.

Purpose of the Study:

• To propose a high-precision polyp segmentation framework, MEFA-Net, to address the limitations of current methods.
• To improve the robustness, generalization, and accuracy of polyp segmentation in endoscopic images.

Main Methods:

• Developed MEFA-Net framework with three modules: Mask Enhancement Module (MEG), Separable Path Attention Enhancement Module (SPAE), and Dynamic Global Attention Pool Module (DGAP).
• MEG module enhances robustness and generalization by masking high-energy regions.
• SPAE module strengthens feature expression to reduce inter-class ambiguity.
• DGAP module addresses intraclass inconsistency by extracting scale, shape, and position invariance.
• Introduced a new evaluation metric, MultiColoScore.

Main Results:

• MEFA-Net significantly improves the accuracy of polyp segmentation.
• The framework demonstrates superior performance compared to current state-of-the-art algorithms across five diverse datasets.
• Quantitative and qualitative evaluations confirmed the effectiveness of MEFA-Net.

Conclusions:

• MEFA-Net offers a robust and accurate solution for polyp segmentation.
• The proposed framework effectively overcomes key challenges in endoscopic image analysis for colorectal cancer detection.
• MEFA-Net represents a significant advancement in automated polyp detection and segmentation technology.