Hybrid Deep Neural Network for Detection of Myocardial Infarction with Electrocardiogram Signals

Accurately diagnosing myocardial infarction (MI) by employing electrocardiogram (ECG) signals is the most rigorous task in detecting and handling heart diseases. MI, colloquially termed as “Heart Attack’, is the most dangerous cardiovascular disease. In some exceptional cases, MI may be asymptomatic; this condition is commonly known as silent heart attack. The mortality rate is very high in the occurrence of silent heart attacks because the patients and doctors do not get enough time to tackle the disease. Therefore, we put forth an automatic detection mechanism for MI employing ECG signals and deep learning models. The proposed methodology incorporates data balancing using SMOTE to address the data imbalance problem, and signal denoising is accomplished using wavelet transform to enhance the quality of the signals. In this work, we have considered 1,23,998 ECG samples captured from “The MIT-BIH arrhythmia database (MITDB)” and “PTB diagnostic database (PTBDB)” to evaluate the proposed methodology. A comparative study of 1D-CNN, LSTM, 1DCNN-LSTM, and 1DCNN-BiLSTM models is performed.1DCNN-BiLSTM achieved efficient results with an accuracy of 99.94%.