FRCNet: Feature Refining and Context-Guided Network for Efficient Polyp Segmentation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Colorectal cancer (CRC) is a prevalent malignancy, with early diagnosis crucial for successful treatment.
• Colonoscopy is the primary screening and diagnostic tool for CRC, requiring accurate polyp identification.
• Polyp segmentation aids clinicians by providing detailed boundary information for diagnosis and treatment planning.

Purpose of the Study:

• To introduce a novel, lightweight feature refining and context-guided network (FRCNet) for real-time polyp segmentation.
• To enhance the accuracy and efficiency of polyp detection in colonoscopic images.
• To provide a robust tool for auxiliary boundary information in clinical analysis.

Main Methods:

• Developed an enhanced context-calibrated module for discriminative feature extraction and long-range spatial dependence.
• Designed a progressive context-aware fusion module to capture multi-scale polyps using multi-range context.
• Implemented a multi-scale pyramid aggregation module for representative feature learning and result refinement.

Main Results:

• FRCNet achieved a 84.9% mIoU and 91.5% mDice score on the Kvasir dataset.
• The model demonstrated superior performance compared to state-of-the-art methods.
• Achieved high accuracy with a small model size (0.78M parameters), indicating efficiency.

Conclusions:

• The proposed FRCNet is effective for real-time polyp segmentation in colonoscopy.
• FRCNet aids in distinguishing polyps from background noise and capturing multi-scale polyps.
• The model offers a promising solution for improving the accuracy and efficiency of colorectal cancer diagnosis.