Automatická detekce různých forem infarktu myokardu s využitím analýzy vektorkadiografických záznamů

Abstract

The work focuses on the automatic detection of different forms of myocardial infarction based on the analysis of vectorcardiographic recordings. The signal was segmented into QRS, ST, and T intervals, followed by the extraction of features describing the shape and spatial properties of the loops. The evaluated parameters included loop volumes, projected areas, directional angles, and mutual relationships between vectors. Machine learning methods were used for classification, particularly the Support Vector Machine (SVM) algorithm and ensemble models. The highest classification accuracy reached 85.48 \% for the task involving two pathological classes and a control group, while for the classification of six pathological classes and a control group, the accuracy was 66.24 \%. The results indicate that the proposed features have potential for myocardial infarction diagnosis; however, increasing the number of classes leads to a decrease in classification performance, especially for mutually similar categories.