Detection of Myocardial Infarction from ECG Signal Through Combining CNN and Bi-LSTM

Myocardia infarction (MI) otherwise known as heart attack, is one of the prime causes of death of individuals worldwide. The electrocardiogram (ECG) signal is clinically used by cardiologists as a screening tool to precisely detect MI. This paper presents a detection approach for MI using 21 time-domain features that are extracted from 12-lead ECG signals. The 12-lead ECG signal facilitates the diversification of features and supports the integrity of machine learning. The features are inserted into a deep learning model based on the combination of a one-dimensional convolutional neural network (1D-CNN) and bidirectional long short-term memory (Bi-LSTM). The 1D-CNN is used for extracting the spatial information of each lead and the Bi-LSTM is used to process the information sequentially in the subsequent leads along with classifying into two distinct classes: healthy and MI. In this paper, the Physikalisch Technische Bundesanstalt (PTB) database is used for evaluation in which 549 patient records are available. The CNN combination with the Bi-LSTM model shows a training accuracy of 99.05% and testing accuracy of 98.50% with 20% data being validated.