Detection of Myocardial Infarction from Electrocardiography Signals with Multiscale Principal Component Analysis and Convolutional Neural Networks

Abstract

Myocardial Infarction is a vital disease that needs to be intervened in a very short time. The analysis of the patient's electrocardiography (ECG) data has an important place in the diagnosis. For this reason, computer aided decision support systems have been used in recent years in order to determine this disease more quickly and accurately. In this study, classification was made using convolutional neural network algorithms on the ECG signals obtained from 61 patients diagnosed with myocardial infarction and 52 healthy individuals. ECG signals are preprocessed with three different filters by applying finite impulse response (FIR) filter, infinite impulse response (IIR) filter and multiscale principal component analysis. According to the results obtained, classification success was achieved with 92.3% accuracy by using the preprocessed signals using multi-scale principal component analysis, and it was seen that more successful classification performance was obtained compared to the classification of the preprocessed signals with the help of FIR, IIR filter.