Topo-UNet: A topology-aware multi-task network for pulmonary vessel segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Radiology

Background:

• Accurate pulmonary vessel segmentation is vital for diagnosing lung diseases.
• Current methods struggle with noisy images, blurred boundaries, fine vessels, and continuity issues.

Purpose of the Study:

• To develop an advanced network, Topo-UNet, for precise pulmonary vessel segmentation.
• To overcome limitations of existing state-of-the-art methods in vessel feature extraction.

Main Methods:

• Proposed Topo-UNet, a topology-aware multi-task network.
• Integrated Bidirectional Slice-wise ConvLSTM (BS-ConvLSTM) for spatial continuity.
• Employed a topology-aware auxiliary task simulating vessel intensity distribution.
• Introduced a joint auxiliary task for vessel refinement.

Main Results:

• Topo-UNet achieved superior performance on CT and CTA datasets compared to SOTA methods.
• Achieved Dice coefficients of 90.78% and 91.91%, and IoU scores of 83.31% and 85.09%.
• Demonstrated enhanced segmentation of fine vessels and improved continuity.

Conclusions:

• Topo-UNet effectively addresses challenges in pulmonary vessel segmentation.
• The proposed methods significantly improve accuracy and continuity, aiding clinical applications.
• The study provides a robust solution for precise pulmonary vessel segmentation.