CAT-WSI: Context-Aware Trajectory Learning for Whole-Slide Breast Pathology Segmentation

Area of Science:

• Digital Pathology
• Medical Image Analysis
• Computational Biology

Background:

• Accurate segmentation of whole-slide images (WSIs) is crucial for computer-aided diagnosis in breast pathology.
• Current patch-wise segmentation methods for WSIs suffer from loss of spatial continuity and global context.
• This limits the performance of automated analysis and treatment planning.

Purpose of the Study:

• To develop a novel framework, CAT-WSI, for context-aware trajectory learning in breast pathology WSI segmentation.
• To overcome the limitations of conventional patch-based methods by preserving long-range spatial dependencies and global context.
• To enhance segmentation accuracy and reliability in digital pathology.

Main Methods:

• Proposed CAT-WSI framework organizes WSI patches into structured transverse trajectories.
• Preserves long-range spatial dependencies and reduces directional bias and boundary fragmentation.
• Augmented trajectory representations with a downsampled whole-slide thumbnail for global-local contextual modeling.

Main Results:

• CAT-WSI demonstrated consistently strong performance across multiple magnification levels.
• Achieved the best overall segmentation results on the CAMELYON16 and Breast-HER2+ datasets.
• Effectively addressed limitations of conventional patch-based WSI segmentation.

Conclusions:

• CAT-WSI offers a significant advancement in breast pathology WSI segmentation.
• The context-aware trajectory learning approach enhances segmentation accuracy and reliability.
• This framework holds promise for improving computer-aided diagnosis and treatment planning in digital pathology.