How to Train Custom Cell Segmentation Models Using Cell-APP.

Area of Science:

• Cell biology
• Microscopy
• Deep learning

Background:

• Deep learning models for cell segmentation require extensive annotated data.
• Manual annotation of microscopy data is time-consuming and limits model development.
• Existing methods struggle with efficient annotation of transmitted-light images.

Purpose of the Study:

• To develop an automated tool, Cell-APP, for annotating training data for transmitted-light cell segmentation.
• To enable the training of high-performance cell segmentation models with reduced manual effort.
• To facilitate the classification of cells based on extracted fluorescence features.

Main Methods:

• Cell-APP utilizes paired transmitted-light (TL) and fluorescence images as input.
• Cell locations are extracted from fluorescence images and used to prompt the μSAM model.
• μSAM generates cell masks on the corresponding TL images.
• Optional: Single-cell features are extracted from fluorescence images for unsupervised classification.

Main Results:

• Cell-APP successfully automates the annotation of training data for TL cell segmentation.
• Trained models demonstrate accurate segmentation and cell-cycle labeling of various cell lines (HeLa, U2OS, HT1080, RPE-1).
• The generated annotations support consistent segmentation for long-time tracking applications.

Conclusions:

• Cell-APP significantly reduces the bottleneck of manual data annotation in cell biology.
• The tool enables the development of robust deep learning models for cell segmentation and analysis.
• Cell-APP is accessible via the Python Package Index with a user-friendly GUI.