Ensemble deep learning prediction of STEMI using ECG signals

Myocardial infarction or heart attack is caused due to atherosclerotic plaque deposition in the coronary arteries thereby occluding the artery, which leads to decrease in blood flow and oxygen supply to the specific regions of the heart muscles. For diagnostic purpose, ECG is used which shows the ST elevation, negative T wave and pathologic Q wave. Classification of myocardial infarction from the normal ECG is handled in this work using the ensemble model of CNN, LSTM and BiLSTM algorithm. The myocardial infarction dataset [10506X188] and normal ECG dataset [4046X188] are retrieved from the PTB Diagnostic ECG Database. The tabular datasets in the size of [14553X191] consisting of abnormal and normal signals and the labels are generated prior to classification. Preprocessing steps involve the signal extraction and signal denoising of both the signal types. The tabular datasets are k-fold cross- validated for training, validation and testing. The split data are trained using CNN, LSTM and BiLSTM network layers individually. The ensemble model, thenceforth, combining all these three networks consecutively is evaluated for the performance in terms of training accuracy 100% and confusion chart for all the four models is also compared.