Detection and Location of Myocardial Infarction from Electrocardiogram Signals Using Median Complexes and Convolutional Neural Networks

Abstract When doctors judge myocardial infarction (MI), they often introduce 12 leads as the basis for judgment. However, repetitive labeling of nonlinear ECG signals is time-consuming and laborious. There a need computer-aided techniques automatic signal analysis. In this paper, we proposed new method based on median complexes convolutional neural networks (CNNs) MI detection location. Median were extracted which retained morphological features MIs. Then, CNN was used to determine whether each lead presented characteristics. Finally, information synthesized realize location Six types recognition performed, including inferior, lateral, anterolateral, anterior, anteroseptal MIs, non-MI. We investigated cross-database performance by method, with models trained local database validated open PTB database. Experimental results showed that yielded F1 scores 84.6% 80.4% test datasets, respectively. The outperformed traditional hand-crafted method. With satisfying generalization performance, may be improved in signals.