VOLTA: an enVironment-aware cOntrastive ceLl represenTation leArning for histopathology.

Area of Science:

• Computational pathology
• Artificial intelligence in medicine
• Biomedical image analysis

Background:

• Accurate cell identification in histopathology images is crucial for clinical oncology.
• Supervised machine learning for this task requires time-consuming manual cell annotations.
• There is a need for automated methods that reduce annotation burden.

Purpose of the Study:

• To propose a self-supervised framework, VOLTA (enVironment-aware cOntrastive cell represenTation learning), for cell representation learning in histopathology images.
• To develop a method that leverages the relationship between cells and their environment.
• To enable cell representation learning without requiring labeled data.

Main Methods:

• Developed a self-supervised learning framework named VOLTA.
• Employed a contrastive learning approach considering cell-environment interactions.
• Trained and validated the model on a large dataset (>800,000 cells) across multiple institutions and six cancer types.

Main Results:

• Demonstrated the effectiveness of VOLTA on ovarian and endometrial cancer datasets.
• Showcased that learned cell representations can identify ovarian cancer histotypes.
• Provided insights linking histopathology and molecular subtypes in endometrial cancer.

Conclusions:

• VOLTA offers a powerful framework for cell representation learning in histopathology without manual annotations.
• The self-supervised approach can empower discoveries, especially in limited sample size scenarios.
• This method has the potential to advance computational pathology and cancer research.