AFANet: Adaptive feature aggregation for polyp segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Deep learning excels in medical image analysis, particularly for polyp segmentation crucial in early colorectal cancer detection.
• Existing algorithms struggle with diverse polyp morphologies and ambiguous boundaries, limiting segmentation accuracy.
• Accurate segmentation is vital for timely diagnosis and effective treatment, impacting colorectal cancer prevalence.

Purpose of the Study:

• To introduce an advanced deep learning model, the Adaptive Feature Aggregation Network (AFANet), for highly accurate colorectal polyp segmentation.
• To address limitations in current methods concerning polyp variability and boundary definition.
• To enhance early detection and treatment strategies for colorectal cancer through improved segmentation.

Main Methods:

• Developed AFANet, incorporating a Multi-modal Balancing Attention Module (MMBA) for refined local feature extraction across foreground, background, and border regions.
• Integrated a Global Context Module (GCM) to leverage encoder-derived global information within the decoder for comprehensive feature analysis.
• Validated AFANet on benchmark datasets (Kvasir-SEG, CVCClinicDB) using Dice and MIoU metrics.

Main Results:

• AFANet achieved high performance metrics: Dice scores of 92.11% and 94.76%, and MIoU scores of 91.07% and 94.54% on Kvasir-SEG and CVCClinicDB, respectively.
• The proposed method demonstrated superior accuracy compared to existing state-of-the-art segmentation algorithms.
• Experimental validation confirmed the effectiveness of both the MMBA and GCM modules in enhancing segmentation.

Conclusions:

• AFANet offers a robust and accurate solution for colorectal polyp segmentation, overcoming challenges posed by polyp diversity and unclear boundaries.
• The model's superior performance indicates its potential to significantly improve diagnostic capabilities in colorectal cancer screening.
• This advancement holds promise for reducing colorectal cancer prevalence through earlier and more precise detection and treatment planning.