CellSeg: a robust, pre-trained nucleus segmentation and pixel quantification software for highly multiplexed
Michael Y Lee1,2,3, Jacob S Bedia4, Salil S Bhate1,2,5
1Department of Microbiology and Immunology, Stanford University School of Medicine, Stanford, CA, 94305, USA.
BMC Bioinformatics
|January 19, 2022
Summary
CellSeg offers a user-friendly, pre-trained solution for nucleus segmentation in multiplexed tissue imaging. This open-source software provides robust cell segmentation, improving quantitative analysis for researchers regardless of programming expertise.
Area of Science:
- Computational Biology
- Bioinformatics
- Image Analysis
Background:
- Quantitative analysis of multiplexed tissue images relies heavily on accurate cellular segmentation.
- Existing segmentation tools often demand manual annotation, extensive training, complex parameter tuning, or advanced programming skills.
- CellSeg addresses these limitations by providing an accessible, pre-trained nucleus segmentation solution.
Purpose of the Study:
- To introduce CellSeg, an open-source software for nucleus segmentation and signal quantification in highly multiplexed tissue images.
- To offer a user-friendly alternative to complex segmentation pipelines, accessible to researchers with varying programming abilities.
- To validate CellSeg's performance against existing state-of-the-art methods.
Main Methods:
- Development of CellSeg using a Mask region-convolutional neural network (R-CNN) architecture.
- Implementation of automated segmentation post-processing steps to enhance cell population resolution.
- Application and validation on diverse multiplexed cancer tissue datasets, including colorectal cancer data from the CO-Detection by indEXing (CODEX) platform.
Main Results:
- CellSeg achieves performance comparable to top algorithms in the 2018 Kaggle Data Challenge, both qualitatively and quantitatively.
- The software demonstrates strong generalization across various multiplexed cancer tissue types.
- Automated post-processing improves the resolution of immune cell populations for downstream single-cell analysis.
- Successful integration into a multiplexed imaging pipeline for accurate cell population identification.
Conclusions:
- CellSeg provides a robust and accessible solution for cell segmentation in highly multiplexed tissue imaging.
- The software empowers biology researchers, irrespective of their programming skill level, to perform quantitative image analysis.
- CellSeg facilitates accurate identification and analysis of validated cell populations within complex tissue microenvironments.
