Detekce infarktu myokardu z vektorkardiografických záznamů s využitím analýzy frekvenční domény

Abstract

Myocardial infarction (MI) is one of the leading causes of death worldwide, and its early detection is crucial for reducing mortality. This thesis focuses on the design of a methodology for the automatic detection of myocardial infarction based on the analysis of vectorcardiographic (VCG) recordings in the frequency domain. Wavelet transform was applied to all three VCG leads for data processing, and both morphological and frequency features were extracted. Feature relevance was analyzed using the MRMR (minimum redundancy maximum relevance) method, based on which four dataset variants were created. Subsequently, the records were classified using machine learning methods including SVM, Random Forest, KNN, and ensemble methods such as AdaBoost, Voting, and Stacking. The highest classification performance was achieved using the ensemble classifier Stacking, which, through the combination of morphological and frequency features, reached an accuracy of 97.54%, a sensitivity of 98.9%, and a specificity of 86.8%, significantly outperforming the other methods. The results demonstrate that the use of frequency features, particularly in combination with morphological features, can substantially contribute to the accurate detection of myocardial infarction.