In this paper, we study features that are not commonly used in clinical practice but may play a role in the automatic detection of acute myocardial infarction (AMI) using a reduced 3-lead ECG system: fragmented QRS in the time domain and intra-QRS in the time-frequency domain. Using chaos theory we reconstruct attractors from ECG and devise geometrical features and two main dynamical invariants: the correlation dimension and the Lyapunov exponent. For validation, we use the Physionet STAFF III dataset. We perform automatic classification using the gradient boosting machine and identify the optimal 3-lead ECG system, achieving promising results: the area under the ROC curve (AUC) is 0.91. It improves the results obtained with the baseline features such as ST-segment elevation and T-wave inversion: AUC 0.85. Finally, we combine new parameters with the baseline features and enhance the final model with previously introduced pseudo-vectorcardiography parameters. The results account for all regions of the heart ischemia: anterior, inferior, and posterior. The proposed automatic algorithm allows the easiest way to determine the first signs of AMI in a patient's ECG based on the input from the minimal number of leads.
|Title of host publication||2021 Computing in Cardiology (CinC)|
|Publication status||Published - 2021|
|Event||2021 Computing in Cardiology (CinC) - Brno, Czech Republic, Brno, Czech Republic|
Duration: 13 Sept 2021 → 15 Sept 2021
|Conference||2021 Computing in Cardiology (CinC)|
|Period||13/09/21 → 15/09/21|