HIGF-Net: Hierarchical information-guided fusion network for polyp segmentation based on transformer and convolution...

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Polyp segmentation is crucial for early colorectal cancer detection via colonoscopy.
• Existing methods struggle with variable polyp shapes, subtle lesion-background differences, and image acquisition interferences, leading to missed polyps and imprecise boundaries.
• Accurate polyp segmentation enhances diagnostic efficiency and patient outcomes.

Purpose of the Study:

• To develop an advanced deep learning model for robust polyp segmentation in colonoscopy images.
• To address limitations of current methods in handling polyp variability and image noise.
• To improve the accuracy and reliability of polyp detection for early colorectal cancer diagnosis.

Main Methods:

• Proposed HIGF-Net, a multi-level fusion network employing a hierarchical guidance strategy.
• Integrated Transformer and CNN encoders to extract both global semantic and local spatial features.
• Utilized a Double-stream structure for inter-layer feature transmission and a Separate Refinement module for boundary enhancement.
• Implemented a Hierarchical Pyramid Fusion module for multi-layer feature merging.

Main Results:

• HIGF-Net demonstrated effective polyp feature mining and lesion identification capabilities.
• The model achieved superior segmentation performance compared to ten existing excellent models across five diverse datasets (Kvasir-SEG, CVC-ClinicDB, ETIS, CVC-300, CVC-ColonDB).
• Experimental results validated the model's learning and generalization abilities.

Conclusions:

• HIGF-Net offers a significant advancement in polyp segmentation for colonoscopy screening.
• The proposed hierarchical guidance and multi-level fusion approach effectively overcomes challenges in polyp detection.
• This model holds promise for improving the accuracy of early colorectal cancer diagnosis.