Cell-APP: A generalizable method for cell annotation and cell-segmentation model training.

Area of Science:

• Biotechnology
• Computational Biology
• Microscopy

Background:

• Deep learning models accelerate microscopy data analysis.
• Manual annotation of cell segmentation data is time-consuming and limits dataset scale.
• Existing methods struggle with diverse cell types and conditions.

Purpose of the Study:

• To develop an automated tool for generating high-quality training data for transmitted-light (TL) cell segmentation.
• To enable the creation of both cell line-specific and general cell segmentation models.
• To overcome limitations of manual annotation in deep learning model development.

Main Methods:

• Cell-APP utilizes paired transmitted-light (TL) and nuclear fluorescence images.
• It extracts cell locations from fluorescence images to prompt deep learning models for cell mask generation.
• Cells are classified as mitotic or nonmitotic based on nuclear features.

Main Results:

• Cell-APP successfully automates the annotation of cell segmentation training data.
• Vision-transformer-based models trained on Cell-APP data achieved high performance.
• Demonstrated the creation of both cell line-specific and multi-cell line segmentation models.

Conclusions:

• Cell-APP significantly reduces the effort required for generating cell segmentation training data.
• The tool empowers researchers to create tailored cell segmentation models.
• Cell-APP provides a scalable approach for developing general cell segmentation models for the research community.