Heart Audio Classification Using Deep Learning

Cardiovascular diseases have been a major cause of deaths amongst the world population. Approximately 17.1 million cases of deaths due to CVD were reported in 2004, this accounts for 29% of the global deaths. Nearly 7.2 million of these cases were related to myocardial heart disease. Thus any approach that could assist in identification of heart disease at preliminary stage is of the utmost importance in improving this elevating index. The work is an advancement in this domain by providing an approach for classification of heart audio samples using some state of the art deep learning techniques. It aims at implementing some existing segmentation techniques along with feature engineering in the audio domain. It also involves a comparative analysis of some existing machine learning algorithms like Support Vector Machine (SVM), Gradient boosting trees, eXtreme Gradient Boosting (XGBoost) along with the deep learning techniques like Convolutional networks(CNN), and recurrent networks like simple Recurrent Neural Networks(RNN), Long Short Term Memory(LSTM). The precision values used to report the model performance comes out to be best for the HybridCNN with about precision of 1 for artifact, 0.906 for normal and 0.859 for murmur category.