Self-supervised machine learning for live cell imagery segmentation.

Area of Science:

• Cell biology
• Bioimage analysis
• Machine learning

Background:

• Accurate cell segmentation is crucial for quantitative analysis of microscopy data.
• Current supervised learning (SL) methods require extensive manual annotation, which is labor-intensive and prone to bias.
• Existing SL models lack robustness and broad applicability in cell biology.

Purpose of the Study:

• To develop an automated cell segmentation method that bypasses the need for human-annotated labels.
• To introduce a self-supervised learning (SSL) approach for robust and unbiased cell and background segmentation.
• To provide a universally applicable segmentation tool for the cell biology community.

Main Methods:

• A novel self-supervised learning (SSL) algorithm was developed.
• The algorithm leverages cellular motion between consecutive microscopy images for self-training.
• No adjustable parameters or curated imagery are required for training.

Main Results:

• The SSL method achieved accurate cell and background segmentation without manual pre-processing.
• The approach demonstrated independence from optical modality and outperformed contemporary supervised learning methods.
• The automated process eliminated user variability and bias in segmentation.

Conclusions:

• This self-supervised learning algorithm offers a first-of-its-kind solution for automated cell segmentation.
• The method is highly accurate, unbiased, and adaptable to user-specific data and imaging conditions.
• It presents an ideal, broadly applicable segmentation tool for cell biology research.