ITMO scientists have developed a new method for automatically calculating the concentration of microalgae cells in solution from microscopic images. Unlike popular neural network approaches, this method is based on classical computer vision algorithms and does not require data collection for training. The code runs on a standard computer and calculates the concentration of cells in a sample in just 30 seconds instead of the usual 30 minutes. At the same time, the accuracy of the data is the same as with manual calculation. The tool will significantly speed up environmental research and increase its accuracy.
Microalgae are actively used in the food industry, cosmetology, pharmaceuticals, water treatment and as alternative fuel sources. Such raw materials are safe for humans and the environment. For example, ITMO ecologists are studying in detail the possibilities of using a type of microalgae in these areas — Chlorella vulgaris.
To grow microalgae on an industrial scale, it is necessary to determine the optimal conditions and environment for their cultivation. One of the most important indicators of the effectiveness of the process of cell growth and division is their concentration. This marker reflects how many cells are in one milliliter or gram of a substance, allows you to compare different samples and determine the causes of an increase or decrease in the number of cells. It is calculated manually according to a technique developed back in the 20th century using a Goryaev counting chamber, a slide glass with markings.
"This method of counting takes a lot of effort and time. At high concentrations of cells, it takes up to 30 minutes to analyze one sample, and more than 10 samples usually need to be processed per day! All this time, the laboratory technician sits at the microscope, continuously and step-by-step working with the sectors of the counting chamber, which often reduces its concentration and errors occur. The only way to make this process faster, more accurate and cheaper is to use tools to automate it," says the author of the study, a graduate student at the ITMO Faculty of Environmental Technologies. Ivan Morshchinin.
There are dozens of similar tools on the market right now. However, the vast majority are neural networks that require large amounts of data for training, special user training, and work only on expensive equipment. In addition, such AI only determines the total number of cells in a sample, and their concentration has to be calculated manually.
ITMO scientists have developed a method for automatically calculating the concentration of cells in samples from images using computer vision technology, which works on a conventional computer. The user uploads images of cells from the microscope to a special folder and runs the code. First, the photos are processed using algorithms: the color spectrum of the images is shifted towards green, the image is converted to black and white and blurred. This is necessary to "clean" the images from the impurities of the solution. Then the Hough circle algorithm is "turned on" to find and distinguish the circles in the image. Since the microalgae cells are round, the algorithm literally "circles" them and separates them from each other. At the last stage, the selected cells are numbered, counted, and their concentration in the sample is determined using a special formula.
"A scientist can always double-check the image processing quality. If the algorithm has poorly selected cells in the photo, you can reconfigure hyperparameters such as magnification of the microscope, volume of the counting camera, resolution of the images, and others, and restart the code. In the case of neural network-based tools, this cannot be done. If at least one parameter of the images changes, for example, the resolution becomes lower, the AI will not work correctly and it will have to be retrained on new data. Our tool is adaptive — it works with images from any equipment and with any parameters," explains the author of the study, a graduate student at the Faculty of Artificial Intelligence Technologies, a junior researcher at the ITMO Strong Artificial Intelligence in Industry Research Center. Julia Borisova.
Algorithms process one "batch" of images in 30 seconds instead of the usual 30 minutes, that is, 60 times faster than a human. At the same time, as the control tests showed, the quality of the data obtained remains the same, and the likelihood of errors due to the human factor is reduced. In addition, compared to the popular pre-trained neural network solution based on StarDist, the developed method identifies cells better: 97% accuracy versus 92%.
According to the curator of the study, Candidate of Technical Sciences, associate professor at the Faculty of Environmental Technology Nelly Molodkina, the use of the new method significantly accelerated the process of their work with chlorella microalgae. According to the ecologist, the use of AI and machine learning methods improves the quality of environmental research results, helps reduce the number of errors and inaccuracies, and expands the understanding of the object being studied through new data.
The developers plan to "package" the tool into a user—friendly UX application, even for the untrained user. And also to work with another type of cells, for example, with the same round-shaped blood cells. In the future, this method can be used to work with oval, fusiform, flat and other cells, replacing the Hough algorithm with a similar functionality, but working with a different type of data.
The study was conducted as part of a collaboration between two ITMO faculties — ecotechnology and artificial intelligence technologies. Students from both departments participated in the project: Yulia Borisova, Ivan Morshchinin and Veronika Nazarova. The supervisors were Nikolay Nikitin, Candidate of Technical Sciences, Associate Professor of the Faculty of Artificial Intelligence Technologies, and Nelly Molodkina, Candidate of Technical Sciences, Associate Professor of the Faculty of Environmental Technologies.
Please note that this press release is based on materials provided by the company. AK&M Information Agency shall not be held liable for its contents, nor for the legal and other consequences of its publication.
