Enhancing Whole Slide Image Classification with Discriminative and Contrastive Learning.

Area of Science:

• Digital pathology
• Computational biology
• Medical image analysis

Background:

• Whole slide image (WSI) classification is vital for digital pathology.
• Large WSI sizes and lack of sub-region labels challenge accurate classification.
• Current deep learning methods struggle with informative image representations for robust WSI classification.

Purpose of the Study:

• To develop an improved method for WSI classification.
• To overcome limitations of existing contrastive learning approaches for WSI analysis.
• To enhance the learning of informative image features for robust WSI classification.

Main Methods:

• Incorporated discriminative and contrastive learning techniques for WSI classification.
• Developed a novel approach focusing on WSI-level positive and negative sample construction.
• Selected representative image patches to create WSI-level samples for effective feature learning.

Main Results:

• The proposed method significantly improved WSI classification performance on two datasets.
• Demonstrated superior results compared to state-of-the-art deep learning methods.
• Enabled the learning of informative features that enhanced the robustness of WSI classification.

Conclusions:

• The developed method offers a significant advancement in WSI classification.
• The WSI-level sample construction strategy effectively learns robust image features.
• This approach holds promise for improving diagnostic accuracy in digital pathology.