Detekce ischemie myokardu s využitím analýzy animálních ortogonálních záznamů

Abstract

Ischemic heart disease (IHD) is one of the most common cardiovascular conditions caused by insufficient oxygen supply to the myocardial muscle often due to coronary artery atherosclerosis. IHD includes myocardial infarction, making its accurate and timely detection crucial. This study focuses on detecting IHD from animal-derived vectorcardiographic (VCG) recordings. VCG data were analyzed and divided into two datasets with different numbers of physiological and ischemic records. One dataset was balanced, while the other was imbalanced. A novel detection methodology was developed using animal orthogonal leads, from which 19 VCG features were extracted. Feature relevance analysis reduced this number to the six most relevant features. These features served as input for machine learning models. A total of eight machine learning methods were tested with experimentally set hyperparameters. Results showed that the most accurate method for the first dataset was Linear Discriminant Analysis (LDA) with 73.08% accuracy and an F1 score of 69.57%. For the second dataset, the best-performing method was Support Vector Machine (SVM) with 85.11% accuracy and an F1 score of 90.91%. For the reduced feature dataset, Adaptive Logistic Regression (ALR) performed best, achieving 81.56% accuracy and an F1 score of 88.79%. Despite the fact that rabbit physiology differs significantly from human physiology, especially in the coronary circulation, which is more interconnected in rabbits and therefore makes ischemic heart disease less apparent than in humans, the proposed methods achieved a high detection success rate. This suggests that the applied methods could be relevant for use in clinical practice.