DSNet: A Dual-Stream Framework for Weakly-Supervised Gigapixel Pathology Image Analysis.

Area of Science:

• Digital Pathology
• Computational Biology
• Medical Image Analysis

Background:

• Whole slide images (WSIs) present classification challenges due to gigapixel resolution.
• Patch-wise classification requires precise, often unavailable, patch-level annotations.
• Existing methods struggle with sub-optimal classification using only image-level labels.

Purpose of the Study:

• To develop a novel weakly-supervised framework for WSI classification.
• To address the limitations of patch-level annotation requirements in WSI analysis.
• To improve classification accuracy by integrating multi-level information.

Main Methods:

• A Dual-Stream Network (DSNet) was developed for WSI classification.
• Local information from patches was encoded into latent embedding vectors.
• Regional information was captured using down-sampled WSI thumbnails.
• The framework was trained using only image-level labels.

Main Results:

• The proposed DSNet framework demonstrated superior performance.
• Outperformed state-of-the-art weakly-supervised WSI classification methods.
• Validation was conducted on three large-scale public datasets.

Conclusions:

• Integrating local and regional information effectively enhances WSI classification.
• The developed framework offers a robust solution for weakly-supervised WSI analysis.
• This approach reduces the need for expensive, precise annotations in digital pathology.