Abstract
Cell counting is a vital practice in the maintenance and manipulation of cell cultures. It is a crucial aspect of assessing cell viability and determining proliferation rates, which are integral to maintaining the health and functionality of a culture. Additionally, it is critical for establishing the time of infection in bioreactors and monitoring cell culture response to targeted infection over time. However, when cell counting is performed manually, the time involved can become substantial, particularly when multiple cultures need to be handled in parallel. Automated cell counters, which enable significant time reduction, are commercially available but remain relatively expensive. Here, we present a machine learning (ML) model based on YOLOv4 that is able to perform cell counts with a high accuracy (>95%) for Trypan blue-stained insect cells. Images of two distinctly different cell lines, Trichoplusia ni (High FiveTM; Hi5 cells) and Spodoptera frugiperda (Sf9), were used for training, validation, and testing of the model. The ML model yielded F1 scores of 0.97 and 0.96 for alive and dead cells, respectively, which represents a substantially improved performance over that of other cell counters. Furthermore, the ML model is versatile, as an F1 score of 0.96 was also obtained on images of Trypan blue-stained human embryonic kidney (HEK) cells that the model had not been trained on. Our implementation of the ML model comes with a straightforward user interface and can image in batches, which makes it highly suitable for the evaluation of multiple parallel cultures (e.g. in Design of Experiments). Overall, this approach for accurate classification of cells provides a fast, bias-free alternative to manual counting.