Cellsketch: Simplified Cell Representation for Label-free Cell and Nuclei Segmentation

Area of Science:

• Computational Biology
• Image Analysis
• Microscopy

Background:

• Accurate cell segmentation is crucial for biological research.
• Existing methods often require manual annotation or struggle with specific imaging techniques like DIC.
• Developing automated and generalizable cell segmentation algorithms is an ongoing challenge.

Purpose of the Study:

• To introduce Cellsketch, a novel technique for automated cell and nuclei segmentation.
• To evaluate the performance of Cellsketch on differential interference contrast (DIC) images.
• To demonstrate the potential of Cellsketch as a generalizable cell segmentation tool.

Main Methods:

• Cellsketch generates an RGB mask with simplified cell representations (nuclei, whole-cell, boundaries).
• A generator model is trained using L1 and adversarial loss for RGB mask creation.
• The watershed algorithm is applied to the RGB mask for final cell and nuclei segmentation.

Main Results:

• Cellsketch achieved accurate segmentation of both cells and nuclei from DIC images.
• The method successfully utilized automatically annotated training data.
• The generated segmentation masks effectively represent individual cells and their nuclei.

Conclusions:

• Cellsketch offers an accurate and automated approach for cell and nuclei segmentation in DIC microscopy.
• The technique shows promise for broad applicability in biological image analysis.
• The availability of the code facilitates further research and development in cell segmentation.