A self-supervised contrastive learning approach for whole slide image representation in digital pathology

Area of Science:

• Digital Pathology
• Computational Pathology
• Artificial Intelligence in Medicine

Background:

• Whole slide images (WSIs) present significant challenges for AI-driven analysis due to their large size.
• Existing self-supervised learning (SSL) methods often rely on patch-based approaches, limiting investigation into end-to-end WSI representation.
• Effective computational pathology requires robust image representation for accurate classification and search.

Purpose of the Study:

• To develop a novel self-supervised learning scheme for whole slide image (WSI) representation.
• To investigate the impact of SSL on end-to-end WSI representation for computational pathology.
• To enhance the robustness of WSI representations for classification and search tasks.

Main Methods:

• Proposed a novel self-supervised learning (SSL) scheme utilizing primary site information, diverging from augmentation-based methods.
• Implemented a fully supervised contrastive learning setup to bolster representation robustness.
• Trained and evaluated the model on over 6000 WSIs from The Cancer Genome Atlas (TCGA).

Main Results:

• The proposed architecture demonstrated excellent performance across various primary sites and cancer subtypes.
• Achieved state-of-the-art results on a lung cancer classification task validation set.
• The novel SSL approach proved effective for end-to-end WSI representation.

Conclusions:

• The novel self-supervised learning scheme effectively addresses challenges in whole slide image analysis.
• The method enhances AI-driven classification and search capabilities in digital pathology.
• This approach offers a promising direction for advancing computational pathology research.