UNSEG: unsupervised segmentation of cells and their nuclei in complex tissue samples

Area of Science:

• Computational Biology
• Biomedical Imaging
• Machine Learning

Background:

• Multiplexed imaging enables sub-cellular resolution analysis of tissue microenvironments.
• Accurate cell and sub-cellular compartment segmentation is crucial for quantifying tissue complexity.
• Existing deep learning methods often require extensive training data and struggle with generalization in unsupervised settings.

Purpose of the Study:

• To develop an easy-to-use, unsupervised segmentation method (UNSEG) for accurate cell and sub-cellular compartment identification.
• To achieve deep learning-level performance without requiring any training data.
• To improve generalization to complex tissue morphologies and enable precise molecular localization.

Main Methods:

• Leveraged a Bayesian-like framework combined with nucleus and cell membrane markers for unsupervised segmentation.
• Introduced a perturbed watershed algorithm for stable and accurate segmentation of individual cell nuclei from clusters.
• Validated the method on gastrointestinal tissue datasets, publicly available datasets, and diverse practical scenarios.

Main Results:

• UNSEG demonstrates internal consistency and superior generalization to tissue morphology compared to current deep learning methods.
• The method accurately identifies cytoplasmic compartments and localizes molecules to their correct sub-cellular locations.
• The perturbed watershed algorithm enhances the accuracy of classical watershed for nucleus segmentation.

Conclusions:

• UNSEG provides a powerful, data-efficient solution for sub-cellular segmentation in complex biological tissues.
• The method facilitates more accurate quantification and analysis of multiplexed imaging data.
• UNSEG offers a generalizable and robust tool for researchers studying tissue microenvironments.