Classification of Myocardial Infarction from Multi-Lead ECG Signals Using Hybrid Deep Learning

Myocardial Infarction, commonly known as heart attack is a serious condition that causes severe damage to the myocardium. Majority of the researches were engaged in preprocessing the ECG signals. This involved various steps such as reducing the noise, removing trends, segmenting beats and selecting relevant features before building models and classifying the data using machine learning methods. In this paper, the proposed model is an ensemble of 12 experts with predefined CNN architectures like AlexNet and GoogLeNet using the ECG signals without any complex preprocessing. Hyper parameters are tuned to give an accuracy of 98.78%, further tested with 6 datasets apart from the PTB database which gives an accuracy of an average of 75%. The model is significant compared with two other methods such as 12 lead scalogram collage method with an accuracy of 20.18% and concatenated raw signal method with an accuracy of 60.43%.