Non-invasive quantification of viability in liver spheroids using deep learning

Area of Science:

• * Biomedical imaging
• * Deep learning
• * Drug discovery

Background:

• * Traditional *in vitro* viability assays are destructive, limiting analyses to single endpoints.
• * Evaluating cell viability is crucial for drug discovery, development, and pharmacovigilance.
• * Existing methods hinder real-time monitoring and comprehensive culture analysis.

Purpose of the Study:

• * To introduce Neural Viability Regression (NViR), a deep learning method for non-invasive, real-time cell viability quantification.
• * To demonstrate NViR's adaptability to different spheroid types via a retrainable pipeline.
• * To utilize NViR for predicting Drug-Induced Liver Injury (DILI) in human liver spheroids.

Main Methods:

• * NViR employs deep learning to analyze microscopy images for real-time viability assessment.
• * The framework was developed and validated using liver spheroids.
• * Human liver spheroids were exposed to 108 FDA-approved drugs, with viability monitored over time using NViR.

Main Results:

• * NViR's viability assessments accurately predicted DILI in humans.
• * The non-invasive approach allowed frequent viability evaluations, capturing temporal changes.
• * Structural integrity of cultures was preserved, reducing experimental costs.

Conclusions:

• * NViR provides cost-effective, non-destructive, high-frequency viability assessments.
• * The technology enhances liver safety protocols in drug discovery and development.
• * NViR has the potential to reduce failure rates and costs in pharmaceutical research.