NSU Scientist Creates Coronavirus Detector

Scientists at the Institute of Automation and Electrometry SB RAS, together with ScientificCoin, developed a method for operative diagnostics of the human organism. They created the Healthmonitor gas analyzer that can determine the presence of coronavirus infection in the body with an accuracy of 85%. More than 2,500 people participated in the study. A person exhales into a special tube and exhalation gases enter a spectrometer that provides their spectral parameters. A neural network developed by Alexander Kugaevskikh, Associate Professor at the NSU Information Technologies Department Computer Technologies Section, then compares these parameters with its data and estimates the probability of having the disease. The patient can find out their results in a couple of minutes with a full analysis through e-mail.

Kugaevskikh described how the project began,

Colleagues fr om ScientificCoin contacted me with this project in June 2020. I was interested in both its social significance and complexity. Only a few scientists have analyzed the spectra of a person's exhalation and screening for several diseases has never been conducted. It was clear from our initial data that the standard approach to the analysis of spectra using neural networks does not work. Almost immediately, I had a hypothesis on how to improve the quality of the neural network. Working with Alina Mikhailenko, a student at the Mechanics and Mathematics Department, from June to October, we searched for different combinations for the gas analyzer parameters and the neural network and in parallel collected a sample of the expiratory spectra. It is always difficult to collect medical data because you need confirmed diagnoses. For the purity of data collection, only one apparatus was used. In October, I started working alone and discovered the right path so I was able to finish adjusting the parameters by December. By this time, a sample of several hundred spectra had been collected so I trained the neural network and the result was 85% accuracy. I gave my colleagues the neural network for installation on their device on December 22. They confirmed the accuracy assessment in test operations on January1 so this was a very good New Year’s gift.

According to the neural network developer, exhalation does not depend on gender and age, but changes only due to acquired or congenital human diseases. To verify the results of the technique, a comparison was made between the patient exhalation data obtained on a gas analyzer with the medical data from PCR analyzes of patients with a confirmed coronavirus diagnosis. Statistical data is currently being collected for official clinical trials that are required for the gas analyzer to obtain a medical certificate. The analyzer can also be used to diagnose diseases such as tuberculosis, asthma, diabetes, and lung cancer.

Kugaevskikh reported that, “Certification in Russia takes three years so our expectation is that now the device will be certified in Europe wh ere the epidemic is still strong. During this time, we will develop new neural network architecture for various other diseases and then we will be able to use it in Russia”.