Application of an Algorithm Based on Transformer Neural Network for Classification of Ultra-high Resolution Electrocardiosignals

Electrocardiography (ECG) is an important and widely used method for diagnosing coronary heart disease (CHD), however, the ECG methods used have disadvantages that do not allow detecting markers of CHD in the early stages of development. In this regard, the ultra-high resolution ECG (UHR ECG) method was proposed in order to identify pathological changes in the cardiovascular system not only in the classical areas of electrocardiosignal (ECS) processing, which are considered artifacts. The main feature of CHD observed with the help of UHR ECG is a significant decrease in the level of spectral power density of the high frequency components of the UHR ECS. However, the manifestation of this sign of the development of CHD can be recorded only as a result of comparing the values for an ECS of living study object before the onset of the disease and after. To overcome this disadvantage, it was decided to use neural networks (NN). As part of this study, an algorithm was proposed in which the architecture based on a transformer network allowed processing a large amount of data, highlighting important information for further processing, and convolutional neural network was used to classify the received signals. This made it possible to determine with high accuracy the moment of the onset of CHD on the UHR ECS obtained during experiments on modeling myocardial ischemia in experimental rats.

Hali tarjima qilinmagan