Variance-aware attention U-Net for multi-organ segmentation.

Area of Science:

• Medical image analysis
• Deep learning in radiology
• Computational anatomy

Background:

• Deep learning advances medical image segmentation.
• Challenges remain in multi-organ, small/irregular area, and ambiguous boundary segmentation.

Purpose of the Study:

• To develop a robust and accurate deep learning model for multi-organ segmentation.
• To address limitations in current medical image segmentation techniques.

Main Methods:

• Proposed a variance-aware attention U-Net architecture.
• Integrated a variance-based uncertainty mechanism to assess voxel prediction probability.
• Employed attention mechanisms to aggregate multi-level features and focus on uncertain voxels during training.

Main Results:

• Achieved superior performance on a challenging abdominal multi-organ CT dataset.
• Outperformed existing attention networks in Dice index (DSC), 95% Hausdorff distance (95HD), and average symmetric surface distance (ASSD).

Conclusions:

• The proposed variance-aware attention U-Net offers an accurate and robust solution for multi-organ segmentation.
• This approach shows potential for enhancing various other medical image segmentation applications.