EmbSAM: cell boundary localization and Segment Anything Model for fast images of developing embryos

Area of Science:

• Developmental biology
• Cell biology
• Computational imaging

Background:

• Cellular shape dynamics are crucial for development but challenging to image due to low signal-to-noise ratios in live-cell microscopy.
• Phototoxicity and photobleaching limit high-resolution, long-duration imaging of cell membranes, hindering accurate shape reconstruction.

Purpose of the Study:

• To develop an integrative computational framework, EmbSAM, for accurate 3D cell membrane segmentation from challenging live-cell images.
• To enable quantitative analysis of cell shape dynamics and morphodynamics during early embryonic development.

Main Methods:

• Integration of a deep-learning-based cell boundary localization algorithm with the Segment Anything Model.
• Application to 3D live-cell imaging of Caenorhabditis elegans embryos with high temporal resolution (10 seconds per stack).

Main Results:

• EmbSAM achieved accurate segmentation of cell membranes despite low signal-to-noise ratios.
• Quantitative characterization of cell-division-coupled morphodynamics, including cell position, division timing, and axis reorientation, prior to gastrulation.

Conclusions:

• EmbSAM provides a robust method for analyzing cell shape dynamics in challenging imaging conditions.
• The framework facilitates detailed investigation of morphogenetic processes and cell fate determination during embryogenesis.