Detection of Heart Attack (Myocardial Infarction) Based on the LSTM Recursive Deep Learning and Clinical Features

Abstract

Deciphering a heart attack (Myocardial Infarction) from an electrocardiogram (ECG) presents a formidable challenge. ECG signals vary greatly among individuals and situations, making interpretation complex. While numerous learning algorithms have been suggested to diagnose myocardial infarction using ECG signals, they often rely on extracted features applied to diverse signals, posing a significant limitation. To address this issue, we employ deep learning, which autonomously learns features without requiring explicit feature engineering. Specifically, we utilize a deep learning-based sequence modeling approach known as short-term long-term memory (LSTM) based on recursive neural networks. We gather data from the Physionet site and the PTB ECG diagnostic database, focusing on ECG records from both healthy individuals and those with myocardial infarction. Simulation results demonstrate that the LSTM algorithm achieves impressive accuracy, recall, and f1-score of 99%, 95%, and 92% respectively for myocardial infarction cases, while also exhibiting 88%, 50%, and 64% for healthy individuals, respectively.