deepOrganoid: A brightfield cell viability model for screening matrix-embedded organoids

Area of Science:

• Biotechnology
• Cancer Research
• Drug Discovery

Background:

• High-throughput viability screens are crucial for chemotherapeutic drug development.
• Patient-derived materials and 3D organoid cultures improve translational relevance.
• Image segmentation in heterogeneous 3D cultures presents quantitative analysis challenges.

Purpose of the Study:

• To develop a non-invasive, label-free tool for evaluating organoid viability.
• To overcome limitations of traditional image segmentation in 3D cultures.
• To leverage deep learning for automated analysis of cellular model systems.

Main Methods:

• Development of a regressive deep learning model.
• Training the model on brightfield images of patient-derived organoids.
• Using terminal viability readout (CellTiter-Glo) as training labels.

Main Results:

• Successful implementation of a data-driven, automated deep learning model.
• Generation of a tool capable of evaluating organoid viability without explicit segmentation.
• Demonstrated potential for label-free assessment of cellular response.

Conclusions:

• Deep learning offers a robust solution for quantitative analysis in complex 3D cultures.
• The developed tool provides a non-invasive method for assessing organoid viability.
• This approach enhances the efficiency and automation of drug screening processes.