Self-supervised machine learning for live cell imagery segmentation

Segmenting individual cells from microscopy images is crucial for quantitative biological data extraction. While supervised machine learning (SL) methods have been used, they require extensive pre-processing with human-annotated labels, which is time-consuming, introduces bias, and lacks generalizability. This paper introduces a self-supervised learning (SSL) approach using cellular motion between consecutive images to train a classifier for cell and background segmentation. This method eliminates adjustable parameters, is modality-independent, outperforms existing SL methods, and is fully automated, thus reducing user variability and bias. The SSL algorithm is a novel approach with promising features for broader adoption in cell biology research.

Michael C. Robitaille, Jeff M. Byers, Joseph A. Christodoulides, Marc P. Raphael