Data-efficient and weakly supervised computational pathology on whole-slide images.

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Digital pathology

Background:

• Deep learning for whole-slide images (WSIs) often needs manual annotation or large labeled datasets.
• Existing methods struggle with domain adaptation and lack interpretability.
• Weakly supervised learning offers an alternative but has limitations.

Purpose of the Study:

• To develop an interpretable, data-efficient deep-learning method for WSI analysis requiring only slide-level labels.
• To improve domain adaptation and interpretability in computational pathology.
• To enable accurate classification and localization of diagnostic features in WSIs.

Main Methods:

• Introduced clustering-constrained-attention multiple-instance learning (CLAM), a weakly supervised deep-learning approach.
• Utilized attention-based learning to identify diagnostically valuable subregions within WSIs.
• Incorporated instance-level clustering to refine the feature space and constrain learning.

Main Results:

• CLAM accurately classifies WSIs and localizes morphological features without spatial labels.
• The method outperforms standard weakly supervised classification algorithms.
• Demonstrated adaptability to independent test cohorts, smartphone microscopy, and varying tissue content.

Conclusions:

• CLAM provides an interpretable and data-efficient solution for computational pathology.
• The method enhances classification accuracy and feature localization in WSIs.
• CLAM shows significant potential for broad application in digital pathology and medical imaging.