SAAVY machine learning program can determine 3D culture viability without harming cells

Cells are either alive or dead. However, in tissue or 3D cultures comprised of many cells, the question is not whether any specific cell is alive, but rather what percentage of cells are still alive, a parameter known as viability.. Many methods that make this viability calculation cannot be used for 3D systems or require killing the cells to obtain the results, preventing longitudinal assays that continuously monitor systems as new chemicals are introduced. Human observation remains a common alternative in the biological and biomedical sciences but has its own drawbacks.

Trettner et al. developed Segmentation Algorithm to Assess the ViabilitY (SAAVY), a machine-learning-based image processing algorithm that can quickly assess viability in 3D culture systems, and demonstrated its use in a longitudinal assay.

To determine viability, the researchers considered the shapes and transparencies of 3D cell systems called pancreatic ductal adenocarcinoma spheroids, which, when healthy, are transparent and circular but gradually lose their shape and turn opaque as the cells comprising the spheroid die.

“We were trying to validate a different technology, and we were having reproducibility issues in assessing spheroid viability,” said author Andrea Armani. “When we approached a colleague about it, they acknowledged that automated spheroid analysis is a common issue, causing many researchers to use manual methods for analyzing the results. Given the recent advances in machine learning, we decided to solve this problem by translating what tissue culture experts see into a quantifiable metric.”

They trained their program to identify the spheroids’ shapes and transparency to score the culture’s viability. They tested SAAVY against a pair of human experts, and it performed similarly while reducing the analysis time by 97%.

Source: “Nondestructive, quantitative viability analysis of 3D tissue cultures using machine learning image segmentation,” by Kylie J. Trettner, Jeremy Hsieh, Weikun Xiao, Jerry S.H. Lee, and Andrea M. Armani, APL Bioengineering (2024). The article can be accessed at https://doi.org/10.1063/5.0189222 .