Myocardial Infarction Detection Using Deep Learning and Ensemble Technique from ECG Signals

Abstract

Automatic and accurate prognosis of myocardial infarction (MI) from electrocardiogram (ECG) signals is a very challenging task for the diagnosis and treatment of heart diseases. Hence, we have proposed a hybrid convolutional neural network—long short-term memory network (CNN-LSTM) deep learning model for accurate and automatic prediction of myocardial infarction using ECG dataset. The total 14552 ECG beats from “PTB diagnostic database” are employed for validation of the model performance. The ECG beat time interval and its gradient value ID are directly considered as the feature and given as the input to the proposed model. The used data is unbalanced class data, hence synthetic minority oversampling technique (SMOTE) & Tomek link data sampling techniques are used for balancing the data classes. The model performance was verified using six types of evaluation metrics and compared the result with state-of-the-art method. The experimentation was performed using CNN and CNN + LSTM model on both imbalance and balance data sample, and the highest accuracy achieved is 99.8% using ensemble technique on balanced dataset.