A Computationally Efficient Hybrid Architecture for Accurate Myocardial Infarction Detection and Localization in ECG Data

Abstract

Myocardial A cardiac event, or myocardial infarction (MI), happens when the blood stream to a part of the heart tissue is blocked. In a lot of cases, MI shows up with mild or unusual symptoms like tiredness, indigestion, or it might not show any symptoms at all until a lot of damage has been done. In extreme cases, it can cause abrupt hemodynamic collapse and even death. Most cases of MI have a connection to initial coronary artery disease (CAD), which is still one of the top causes of death in the US. Early detection and accurate localization of myocardial infarction are essential for averting additional cardiac complications and enhancing patient outcomes. Current methods have created lightweight and effective machine learning-based systems their use features from single-lead ECG signals to automatically find different types of MI. An autonomous encoder and decoder network is used to learn representations on a that store the most important features of the cardiac rhythm signal in order to improve feature quality. Then, a mix of algorithms used for machine learning, such as Random Forest as well as XGBoost, is used to make the classification more accurate. These systems have done very well, with a reported accuracy of 96.75%. To improve detection performance even more, the proposed framework adds a model that makes very expressive along with overfitted representations for ECG data. This model uses a DNN, or deep neural network, that was purposely trained in an overtime mode so that it can pick up even the smallest changes on the PTB-XL heart attack study data. This method greatly improves predictive performance by learning complex and relationships that are not linear in the data. When tested, the proposed model gets 99.90% accuracy on test data and almost perfect scores on other important performance metrics. This shows that it has the potential to be very reliable for MI finding and localization.