Heart Disease Recognition from ECG Signal Using Deep Learning
Automated detection method for multiple heart diseases using Electrocardiogram (ECG) signal is proposed which is very much necessary for precise prediction of any heart disease at an early stage. The research work discusses an approach for detecting various cardiovascular diseases (CVDs) based on deep learning approach. The use of one dimensional Convolutional Neural Network (CNN) modelling for predicting any of CVDs is a relatively new approach. Before training a neural network with ECG data appropriate preprocessing is performed. An adaptive method, Empirical Mode Decomposition is used for removing high frequency as well as low frequency noise components. The datasets of multiple heart diseases such as atrial fibrillation (AF), myocardial infarction (MI), congestive heart failure (CHF) and normal are sampled to address the class imbalance problem. The heart beats are extracted from the ECG signal by identifying R peaks using Discrete Wavelet Transform method, the segmented heart beats are given as input to the CNN to classify among multiple diseased classes. The model described in this work has a testing accuracy of 86% for classification of the diseases CHF and MI and 81% for CHF vs normal heartbeat. This model can be further extended to identify more diseases.
Keywords: Empirical Mode Decomposition (EMD), 1D- Convolutional Neural Network (CNN), Deep learning, Intrinsic Mode Decomposition (IMF), Cardiovascular diseases.