The use of artificial intelligence in the analysis of ultra-high resolution electrocardiograph data

The purpose of this study is to demonstrate the possibility of classifying measurement data of a small-sized block for logging, registration and recording electrocardiosignals (ECS) using the ultra-high resolution electrocardiography (UHR ECG) method, which is being developed in Laboratory 235 of the Institute for Analytical Instrumentation of the Russian Academy of Sciences (IAI RAS). To classify the results of electrocardiographic measurements, a 4-layer convolutional neural network (NN) was used, trained and tested on a database consisting of 90 records of UHR ECS obtained during experiments modeling acute myocardial ischemia in experimental rats. Each experiment included three stages — "stabilization", "ischemia" and "reperfusion", characterized by different morphology and spectral characteristics of the recorded signals. The task was to automatically estimate the probability of each cardiac cycle belonging to one of the 4 classes corresponding to the three main stages of the experiments, as well as the transitional stage between the stages of "stabilization" and "ischemia". To assess the quality of the classification, a learning curve of the modeland a time dependence of the probability assessment of each cardiac cycle belonging to one of the 4 classes of an UHR ECS recordwere plotted.