Researchers use microscopy images in many ways to keep track of moving cells or microscopic organisms. Now researchers have developed an algorithm that makes this task much easier. The algorithm optimises the analysis of images of a frequently studied worm.
For centuries, scientists have sat hunched over their microscopes studying many tiny phenomena, from cells to the movement of minuscule organisms.
However, researchers have always faced the obstacle of how to keep track of the objects that whizz across their field of vision, making it hard to quantify or generally monitor what is what. These objects include sperm cells and nematodes – small worms used in thousands of studies of various diseases.
In recent decades, image analysis has helped researchers to analyse the number of cells or animals in their samples and how they move, but the researchers are often challenged because the computer algorithms have required low density to analyse the images and keep track of both number and movement.
However, this challenge may soon be overcome, since researchers have developed a new algorithm that quantifies cells and detects cell movement in even very dense samples.
“Our approach enables us much more easily to investigate the effectiveness of drug candidates in experiments with nematodes or how much sperm cells move in dense samples. The algorithm leverages the potential of computer vision and is a proof of concept for performing accurate image analysis on samples with a much higher density of cells than is possible today,” explains a researcher involved in the study, Julius Kirkegaard, Assistant Professor, Biocomplexity, Niels Bohr Institute, University of Copenhagen.
The research has been published in Communications Biology.
Image analysis is essential in research
Researchers in neurology often expose the Caenorhabditis elegans nematode to various chemicals to determine how this affects their neurons. Many drug candidates for diseases of the nervous system are also initially studied in experiments with nematodes.
However, when researchers study the effect of this exposure, they often have to reduce the density of their samples of nematodes so that they do not overlap under the microscope – otherwise, the computer-based detection models that analyse their movements cannot keep track.
The smart thing about using algorithms to analyse images and videos of nematodes in general is that the researchers can obtain very accurate numbers for analysis and statistics.
The algorithm Julius Kirkegaard has developed in collaboration with Albert Alonso enables experiments to be analysed even at high density and thus enables very large data sets to be rapidly created.
“Our algorithm for analysing images and videos of nematodes enables us to identify very precisely how much they move in an image and how each individual organism acts. Current methods are either limited to very low density or ignore problems related to overlap between organisms,” says Julius Kirkegaard.
Two innovations make the model superior
The algorithm the researchers developed combines classical biophysics and modern machine learning and broadly contains two innovations that previous algorithms did not have.
First, the algorithm enables a formula to be inserted for how the observed bodies move.
For example, researchers have already established clear mathematical formulas for how a nematode moves, and these can be entered into the algorithm. This minimises the need for large data sets, since the algorithm does not have to learn these movement patterns from the data but is programmed instead to understand the task based on the principles of physics.
The second innovation involves managing how the observed bodies overlap each other.
Most classical and deep learning–based algorithms work by naming the pixels in the image and assigning the pixels to the individual organisms.
However, this becomes a problem when two organisms overlap and one pixel thereby actually relates to two or more organisms.
The researchers solved this problem in the new algorithm by enabling the algorithm to encode a virtual fingerprint for each organism based on its movement patterns. This enables the algorithm to monitor each organism, even when they overlap in the image.
Making impossible research possible
Julius Kirkegaard says that the new algorithm can keep track of overlapping organisms much better than has been possible so far.
However, the algorithm has limitations that prevent it from being immediately used by everyone who works with moving bodies under a microscope.
The biggest problem is that the algorithm requires a mathematical formula for the movement of the bodies under the microscope.
“So far, the algorithm is therefore a specialist, not a generalist. Right now it is state of the art for researchers studying nematodes, and they can replace the formula for the movement of nematodes with a formula for the movement of the bodies they are studying, but this requires that these formulas be known and applicable . A few innovations are still required before the algorithm becomes equally usable for everyone,” explains Julius Kirkegaard.
Nevertheless, the researchers also showed that the algorithm enables them to analyse videos of nematodes at a much higher density than was possible before.
“We have gotten people who work with nematodes to perform experiments at a much higher density than they usually do, and this demonstrates that this previously impossible analysis is now possible,” concludes Julius Kirkegaard.