Vladimir Andryushchenko, a postgraduate student at the TSU Institute of Applied Mathematics and Computer Science, develops methods and algorithms that will help to determine and predict changes in the patient's condition based on medical signals. He is creating a large library of medical data required for machine learning by a computer model that will verify diseases using electrocardiogram (EKG) signals.
- A significant part of the data used for diagnostics is still in analog form, which significantly reduces the possibility of fully analyzing them, - says Vladimir Andryushchenko. - This problem can be solved by switching to digital. The accumulated data for one patient or group can contain useful information not only about the current state of health but also about the beginning of critical changes in the human body.
As he notes, there are now examples of successful application of machine learning methods to detect a specific disease or class of diseases, but there is no universal way to detect a wide range of diseases. To create such an algorithm, we need a huge training sample, which will have a large number of patterns - repeating elements characteristic of each class of disease.
The task of the project is to form such a data library. Along with the patterns that are registered in various diseases, the library will include a large array of EKG results obtained from examining healthy people. Learning on this sample will help artificial intelligence to separate the norm and pathology, to determine the type of heart failure.
Under the new project, the young scientist creates algorithms for training a computer model for classifying EKG signals, which help to build an effective neural network while, avoiding excessive complexity. The tasks also include developing of a method for identifying patterns of electrical signals from the heart of a healthy and a sick person.
Subsequently, by analyzing these patterns, artificial intelligence will be able to identify different types of cardiopathologies, for example, diagnose conditions associated with cardiac arrhythmias - sinus arrhythmia, sinus tachycardia, extrasystole, and others.
Along with this, algorithms for analyzing the dynamics of changes in the patient's EKG data at different times will be created. That will make it possible to identify significant changes even before the appearance of visible symptoms. At the final stage, a prototype of a program for automatic EKG analysis will be proposed and the results will be tested on real data from medical diagnostics.