A Novel Approach for the Detection of Cardiovascular Abnormalities from Electrocardiogram and Phonocardiogram Signals Using Combined CNN-LSTM Techniques

Heart diseases account for 30 percent of the fatalities worldwide.Early intervention and detection of cardiovascular abnormalities can prevent such fatalities.The current research proposes a novel approach combining Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) for the prediction of abnormalities in the functioning of the human heart.The machine learning model is used to detect abnormalities from ECG and PCG signals.Two prominent datasets namely Physionet 2016 and Physionet 2017 have been used in this research for training and testing the developed machine learning model.Empirical Mode Decomposition has been used for preprocessing the heart sound signals and ECG signals.A signal can be broken down into its fundamental oscillatory components, known as intrinsic mode functions (IMFs), using EMD.By comparing the signal to noise ratio value to the raw and filtered PCG signal, one may evaluate the method's effectiveness in reducing noise.Feature extraction is done through the generation of Scalograms of the denoised signals.The scalogram is obtained by continuous wavelet transform (CWT).After this, a hybrid deep learning technique called CNN-LSTM is used for classifying and training the model.The proposed model renders an accuracy of 86% in terms of classifying and detecting abnormalities in the functioning of the human heart.