Hierarchical spatial perception network and SAM-assisted uncertainty suppression for medical image segmentation

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate medical image segmentation is vital for image-guided procedures.
• Anatomical complexity and variability pose significant challenges to current segmentation methods.

Purpose of the Study:

• To develop a novel medical image segmentation framework integrating Hierarchical Spatial Perception network (HSP-Net) and a foundation model-assisted uncertainty suppression (SUS) strategy.
• To enhance the accuracy and robustness of medical image segmentation.

Main Methods:

• Proposed a novel segmentation framework combining HSP-Net with a SUS strategy leveraging the Segment Anything Model (SAM).
• HSP-Net incorporates a Group Pyramid Attention (GPA) module with Channel Pooling Attention (CPA) and Spatial Hierarchical Attention (SHA) blocks.
• The SUS strategy utilizes SAM to identify challenging pixels and reduce prediction uncertainty.

Main Results:

• The proposed framework achieved an average Dice score of 0.8815 and a 95% Hausdorff distance of 7.3867 across five public datasets.
• Demonstrated consistent outperformance compared to U-Net and its variants (e.g., UNeXt, UTNet).
• Showcased robust generalization capabilities across multimodal medical images.

Conclusions:

• The integrated framework effectively combines semantic feature learning with foundation model-assisted refinement for superior medical image segmentation.
• The novel approach offers a promising solution for accurate and reliable segmentation in clinical applications.
• The source code is publicly available for further research and development.