Roman Kostyuchenko, a bachelor's degree student at the Faculty of Applied Mathematics, Informatics, and Mechanics at the Voronezh State University, wrote a program for automatic identification of pneumonia types from X-ray images. This was reported by the VSU press service on Thursday, July 9.
The development has become especially relevant during the spread of the new coronavirus infection. Today, when diagnosing viral pneumonia, they commonly use the method of computed tomography (CT) which is not without its drawbacks. The most important of them is the high radiation load on the body. In addition, there are possible errors of the radiologist or the problem of excess weight which does not allow some patients to be examined.
‘The topic of my graduate bachelor's work is ‘Development of a Pneumonia Recognition System on X-ray Images’. I chose it together with the research supervisor. I had to develop a system that would diagnose the presence or absence of pneumonia on X-rays of the lungs and, if it was present, would determine the type - bacterial or viral. In the course of creating the system, I considered the general tasks of medical diagnostics, the principle of operation of neural networks, studied the experience of their usage in other medical projects. As a result, I managed to create a system based on the technology of ultra-precise neural networks that recognizes either the absence of pneumonia or its type from X-ray images. It is yet too soon to talk about implementation since such medical decision support systems require a long improvement because this is a very big responsibility. I would like to find new solutions and technologies to improve the accuracy of forecasting as part of my master's degree work,’ said Roman.
The research supervisor of the project, Professor of the Department of Mathematical Methods of Operations Research Irina Kashirina emphasized that the research should be continued with increasing the measurement accuracy:
‘Of course, we are not the first to undertake this task. However, until it is solved, we have a chance to make the development more accurate and therefore more efficient. Artificial intelligence can see slightly better than the eye of a radiologist. Right now everyone with suspected viral pneumonia is sent for a CT scan. If the diagnosis is confirmed, the examination is done several times to analyze the dynamics of the disease. But this is a huge radiation load on the body - hundreds of times higher than with conventional digital X-ray. The main goal of our study was to learn how to identify viral pneumonia on X-rays and distinguish it from the bacterial type. So far we are only at the beginning of the path: the accuracy of the created algorithm is over 80%. We are planning to increase it to 95% in the near future.