3D cell tracking software enables researchers to track the development of cells in tissues automatically. Using neural networks, these programmes recognise where cells are moving and where new cells have originated in a series of microscope images. However, this method can be inaccurate in tissues with large numbers of cells, which requires time-consuming manual correction.

‘In our group, we are investigating the development of intestinal organoids, which are pieces of intestinal tissue that we grow in the laboratory’, says Max Betjes, a PhD student and biophysicist at the AMOLF physics research institute. ‘Tracking is very difficult in this case because the intestinal cells are very close together and divide very quickly. The manual correction work involved can easily account for half of a PhD programme.’ This is why Betjes developed a new algorithm that not only tracks cells more effectively, but also enables the underlying neural network to highlight potential errors. This makes it much easier to identify which parts of the cell tracking need correcting. Betjes explains, ‘In this way, our algorithm increases the throughput of experiments with organoids.’

Political philosophy

The fact that neural networks indicate a certain probability that their prediction is correct or incorrect is nothing new. A classic example is a neural network presented with a picture of a dog or a cat that has to determine which it is. In doing so, the network establishes an underlying probability distribution to show how confident it is in its answer. ‘But our question is much more complicated than that single decision’, says Betjes. ‘With organoids, the neural network has to look at thousands of cells at once. That is often still too difficult.’

This is why the researchers’ algorithm instructs the neural network to examine each cell in an image individually and then combines all the results within a statistical framework. This results in a margin of error that indicates the certainty with which the cell tracking prediction was made. Betjes: ‘The innovation of our algorithm lies in the rigorous statistical approach we use to achieve this.’

Although Betjes was mainly inspired by statistical physics, he also refers to political philosophy as an analogy, where he encountered a similar statistical approach. ‘Suppose you have many experts who all say something about reality. How do you combine all those opinions into a single consensus? We also use the underlying statistics that enable you to do this in our algorithm.’

Cell tracking challenge

The researchers demonstrated the effectiveness of their algorithm by testing it on intestinal organoid data from their own laboratory. They also used publicly available data from early-stage mouse embryos and from C. elegans, a worm whose embryonic development has been known since the late 1990s to be identical and to have been tracked manually.

Betjes even participated in the Cell Tracking Challenge’, a competition that ranks cell tracking algorithm data analysis by comparing it to manually tracked C. elegans data. He came first. ‘While this only shows how accurately our algorithm made the predictions, what is more important to us is that the algorithm correctly indicates the certainty with which it made that prediction.’

Getting started

To further validate their algorithm, Betjes and his colleagues from the groups of Sander Tans and Jeroen van Zon are encouraging other groups to start using this method. To this end, they have built a website where other researchers can test the algorithm on their data without downloading the software.

‘Our algorithm will probably be one of the best for tracking cells in complex tissues over the next two years’, says Betjes. ‘That’s why we are now focusing on reaching people who could also benefit from this method.’

Anyone interested can try out the algorithm at OrganoidTracker.org.