Transformer-based spatial-temporal detection of apoptotic cell death in live-cell imaging.

Area of Science:

• Cell biology
• Computational biology
• Bioimaging

Background:

• Intravital microscopy enables live-cell imaging of spatial-temporal cell dynamics.
• Quantifying cellular processes like apoptosis in microscopy data remains challenging.
• Apoptosis is critical for tissue homeostasis and host defense.

Purpose of the Study:

• To develop a robust computational method for detecting apoptosis in live microscopy timelapses.
• To create a deep learning system for accurate apoptosis detection and quantification.
• To overcome limitations in analyzing complex intravital microscopy data.

Main Methods:

• Developed ADeS, a deep learning-based apoptosis detection system using activity recognition principles.
• Trained ADeS on over 10,000 in vitro and in vivo apoptotic instances.
• Validated ADeS across diverse imaging modalities, cell types, and staining techniques.

Main Results:

• ADeS achieved >98% classification accuracy, surpassing state-of-the-art methods.
• The system accurately detects the location and duration of multiple apoptotic events in full timelapses.
• ADeS demonstrated superior performance compared to human analysis.

Conclusions:

• ADeS is the first method for robust apoptosis detection in microscopy timelapses.
• The system offers accurate quantification of cell survival and tissue damage.
• ADeS is a valuable tool for analyzing intravital microscopy data and understanding apoptosis regulation.