MGF-net: Multi-channel group fusion enhancing boundary attention for polyp segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Colonic polyps are common neoplastic lesions during colorectal cancer screening.
• Early detection and removal of polyps are vital for preventing cancer and reducing mortality.

Purpose of the Study:

• Develop a high-precision intelligent polyp segmentation network to enhance polyp screening during colonoscopies.
• Address the need for improved automated polyp detection systems.

Main Methods:

• Utilized ResNet50 as the backbone with a multi-channel grouping fusion encoding module for feature extraction.
• Employed receptive field and grouping fusion modules to capture multi-scale and salient features.
• Introduced an enhanced boundary weight attention module and self-attention mechanism for refined boundary segmentation.

Main Results:

• MGF-Net demonstrated significantly improved segmentation accuracy compared to mainstream networks on five public datasets.
• Contrast experiments and hypothesis testing confirmed the statistical significance of the results.

Conclusions:

• The proposed MGF-Net network offers superior performance over existing polyp segmentation models.
• MGF-Net presents a promising advancement for intelligent polyp detection in colonoscopy screenings.
• The MGF-Net model code is publicly available for further research and development.