Myocardial infarction detection using ECG signal applying deep learning techniques - ConvNet, VGG16, InceptionV3 and MobileNet

Abstract

Due to our unhealthy diets and the consumption of enhanced cholesterol in our daily lives, our health has become vulnerable and at risk of different types of cardiac diseases. The most common of them is Myocardial Infarction (MI), also known as Heart Attack. Myocardial Infarction takes place because of sudden blockage of blood flow in one’s heart. Without sufficient blood flow, one’s heart muscles cannot get the nourishment and oxygen that they need to function appropriately, which causes irreversible damage to the heart tissues. However, early detection and treatment of a Myocardial infarction can decrease the risk of heart damage and increase the rate of survival. As a diagnostic tool, the Electrocardiogram (ECG) is one of the most popular to diagnose various cardiovascular illnesses, including Myocardial Infarction (MI). The ECG captures the heart’s electrical activity and these signals can be utilized to diagnose irregular cardiac rhythms. Because of the intensity and duration of ECG signals, manual ECG signal diagnosis is prone to errors and is neither sensitive nor specific for MI diagnosis when used alone. Therefore, this research proposes a novel approach of detecting Myocardial Infarction (MI), using deep learning techniques. It includes ConvNet model as well as other popular transfer learning models like MobileNet, VGG16 and InceptionV3 which uses 12-lead ECG signals as input. The trained model with the proposed ConvNet and MobileNet architecture have shown exceptionally promising accuracy in MI detection compared to VGG16 and InceptionV3. The performance of the proposed models are measured using Confusion matrix , Precision score, F1-score, Recall score and ROC curve. Our average accuracy is 97.50 percent which is acquired by using MobileNet. Also, the Convnet model shows promising result. Thereby, we can say that the suggested model can deliver high MI detection performance in wearable technologies and intensive care units.