Morphological and Temporal ECG Features for Myocardial Infarction Detection Using Support Vector Machines

Abstract

Myocardial infarction is a leading cause of death worldwide. A 12-lead electrocardiogram (ECG) recording is commonly performed to diagnose this pathology. In this paper, we explored temporal and morphological features extracted from multi-lead ECG signals to classify subjects from the PTB Diagnostic ECG database into healthy control and myocardial infarction using a support vector machine binary classifier. After delineating the 12-lead ECG signals with a wavelet transform-based method, a unique set of characteristic points was obtained for the ECG leads by suppressing outliers and by taking the average of the remaining points. Then, mathematical operations (average, standard deviation, skewness, etc.) performed to the P wave duration, QRS complex duration, ST-T complex, QT interval, T wave duration and RR interval were used as temporal features, and mathematical operations performed to ECG signals bounded by the P wave, QRS complex, ST-T complex and QT interval were used as morphological features. A 10-fold Monte Carlo cross-validation was employed to analyze the reproducibility of the classification results by randomly splitting the dataset into training (70%) and test (30%) sets with balanced classes. Mean classification accuracies above 93% were achieved when the SVM classifier uses only temporal ECG features, only morphological ECG features, and both temporal and morphological ECG features. The best classification performance was achieved when temporal and morphological ECG features are jointly considered by the binary SVM classifier (accuracy 96.67%, error rate 3.33%, sensitivity 97.33% and specificity 96.00%).