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

Abstract

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