Siberian Federal University

This development can be used in medicine for diagnosing and predicting socially significant diseases, for example, the risks of complications of myocardial infarction and oncological diseases. According to the scientists, the method is versatile enough to be used wherever a quick and reasonable analysis of a large amount of data is required, for example, in production for sorting parts, and even for solving tax issues.

One of the basic issues of modern machine learning is the creation of explainable artificial intelligence, the solutions of which will be advisory in nature — the so-called decision support system. The algorithm of this type of AI is completely transparent.

“Modern machine learning algorithms, for example, neural networks, most often are black boxes. According to the input data, they form a certain result that we either trust or do not. However, in medicine or in the taxation system, the user needs an explanation of why the machine came to this or that conclusion, because the consequences of a negative AI decision can be devastating. This is especially true in the field of diagnosing diseases or modeling possible complications for the health of patients,” informed Igor Masich, research leader, professor at the Department of Information Systems, School of Space and Information Technologies, SibFU.

In case AI is trained by a teacher, it is given cases with labels. Based on these initial data, the AI can find logical patterns and build a classifier consisting of logical rules. The AI correlates with these rules each new case that requires consideration.

The scientists have already found out how the new algorithm will behave in predicting complications in patients with myocardial infarction. The task was to predict the condition of such patients and adjust their individual treatment.

“You can take the results of primary tests and examinations, for example, blood tests, heart ultrasound and coronary angiography, electrocardiogram, etc. and upload them into a system that, according to the given rules, will check everything, correlate, and draw conclusions. The physician can interpret these findings and use them to prescribe individual therapy,” said the scientist.

Thus, the machine will assist in predicting not only the main complications in the post-infarction period, but also warn the patient about the possibility of a repeat crisis, or even be proactive in risk groups and predict the likelihood of a heart attack.

Cancer diagnosis is another medical challenge that the new algorithm is already learning to handle. The SibFU scientists instructed the AI to solve a genetic problem, namely, to study gene expression in order to predict the development of oncological diseases. The machine assistant had to study a huge amount of data, and that was really challenging. Maria Bartosh, a student of the School of Space and Information Technologies, SibFU, managed to identify several genes with the help of the new algorithm, which might be responsible for the development of cancer in the human body. The machine found them, analyzing individual patient data and correlating them with thresholds.

The next step was to apply logical data analysis to more complex and informal clustering problems. The scientists have developed a tool for completely transparent grouping of objects into clusters using concepts close to human thinking. This tool is also useful for more detailed classification and identification of new classes of objects, for example, the detection of new types of diseases.

In addition, the scientists have provided another application of the developed system. It will help employees of tax authorities to check information about taxpayers, verify the accuracy of tax returns, etc. Such AI abilities will also be useful in production to allocate homogeneous batches of products with increased quality requirements, for example, electronic component products for space applications. Thus, one can quickly and efficiently sort microchips or transistors according to their qualities.