PEA-Net: A progressive edge information aggregation network for vessel segmentation.

Area of Science:

• Medical image analysis
• Computer-assisted diagnosis
• Vascular imaging

Background:

• Accurate automatic vessel segmentation is crucial for diagnosing vascular diseases.
• Challenges include uneven contrast, background noise, and fragmented results from existing methods.
• Current techniques often overlook vessel morphology, leading to segmentation errors.

Purpose of the Study:

• To introduce a novel network, PEA-Net, for improved automatic vessel segmentation.
• To address limitations of existing methods in handling noise and preserving vessel structures.
• To enhance both pixel-level accuracy and topological connectivity in vessel segmentation.

Main Methods:

• Proposed PEA-Net utilizes a dual-stream receptive field encoder (DRE) to capture fine structural features and reduce noise.
• Incorporated a progressive complementary fusion (PCF) module to enhance vessel detection and connectivity by integrating nonsalient information.
• Employed segmentation-edge decoupling enhancement (SDE) modules as decoders for integrating features and refining segmentation.

Main Results:

• PEA-Net demonstrated superior performance in vessel segmentation across multiple datasets.
• The model achieved optimal results in both pixel-level and topology-level evaluation metrics.
• The proposed strategy effectively reduced topological disconnections and improved overall segmentation quality.

Conclusions:

• PEA-Net offers a significant advancement in automatic vessel segmentation for medical imaging.
• The network's architecture effectively handles image noise and preserves intricate vessel structures.
• This method holds promise for more accurate and reliable diagnosis of vascular diseases.