Edge guided bidirectional iterative network in medical image segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Medical image segmentation is challenged by blurred edges, impacting accuracy for complex anatomical structures.
• Current edge-enhanced networks often suffer from unidirectional information flow, limiting performance.

Purpose of the Study:

• To propose a novel edge-guided bidirectional iterative network (EGBINet) for enhanced medical image segmentation.
• To improve segmentation accuracy by enabling bidirectional flow of edge and region information.

Main Methods:

• Developed a cyclic network architecture enabling bidirectional information flow between encoder and decoder.
• Introduced a transformer-based multi-level adaptive collaboration module (TACM) for improved feature fusion.
• Fused edge features with multi-level region features for enhanced feedforward pathways and iterative optimization.

Main Results:

• EGBINet demonstrated significant performance advantages over state-of-the-art methods on multiple medical image segmentation datasets.
• Achieved superior edge preservation and segmentation accuracy for complex anatomical structures.
• Validated the effectiveness of the bidirectional iterative approach and TACM module.

Conclusions:

• The proposed EGBINet effectively addresses limitations of unidirectional flow in medical image segmentation networks.
• Bidirectional information exchange and adaptive feature fusion lead to superior segmentation performance.
• EGBINet shows great potential for clinical applications requiring high-accuracy medical image segmentation.