MFEFNet: Multi-scale feature enhancement and Fusion Network for polyp segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Colorectal cancer prevention relies on early polyp detection.
• Accurate polyp segmentation is crucial for computer-aided diagnosis.
• Existing methods face challenges in feature extraction and fusion for precise segmentation.

Purpose of the Study:

• To propose the Multi-Scale Feature Enhancement and Fusion Network (MFEFNet) for precise polyp segmentation.
• To improve the accuracy and efficiency of polyp detection in medical images.
• To enhance the capabilities of computer-aided systems in preventing colorectal cancer.

Main Methods:

• Utilized ResNet50 as the backbone network with Shift Channel Blocks (SCB) for feature unification.
• Incorporated Feature Enhancement Blocks (FEB) for multi-perspective feature reinforcement.
• Introduced Multi-Scale Feature Fusion Blocks (MSFFB) and Reducing Difference Blocks (RDB) to address semantic gaps.
• Employed Polarized Self-Attention (PSA) and Balancing Attention Modules (BAM) for detailed boundary exploration.

Main Results:

• The MFEFNet demonstrated significant improvements in polyp segmentation accuracy.
• Achieved an average increase of 3.4% in Dice score and 4% in mean intersection over union (mIoU).
• Outperformed over a dozen state-of-the-art methods on five benchmark datasets (Kvasir-SEG, CVC-ClinicDB, CVC300, CVC-ColonDB).

Conclusions:

• The proposed MFEFNet effectively segments polyps with high precision.
• The network's advanced feature enhancement and fusion mechanisms contribute to its superior performance.
• MFEFNet shows strong potential for advancing computer-aided polyp detection and colorectal cancer prevention.