Accurate Classification of Heart Sound Signals for Cardiovascular Disease Diagnosis by Wavelet Analysis and Convolutional Neural Network: Preliminary Results

Abstract

Heart sound (HS) signals contain valuable diagnostic information for detection of heart abnormalities. The early detection of heart abnormalities plays an important role in reducing the mortality rate caused by heart diseases. Auscultation, the process of listening to heart sounds, is the first diagnostic method of heart diseases. This process is highly dependent on the physician expertise, making the diagnosis more of a subjective issue. There is ongoing research to automate heart sound diagnosis. Advances in machine learning have provided an easier, cheaper and objective diagnosis of diseases. Algorithms developed for heart sound classifications rely on several features and the accuracy of a model depends on the feature vector. The advent of deep learning (DL) provides a possible solution to overcome the overwhelming and time-consuming step of feature extraction. Convolutional neural networks (CNN), popular deep network architectures, offer high classification accuracies for 2D images and 1D time series. This study proposes an efficient and highly accurate method for heart sound signal classification. The continuous wavelet transform method is employed to obtain scalogram images. The 2D scalogram images are fed to a deep CNN classifier. Using the heart sound dataset consisting of 4 abnormal and 1 normal heart sound subsets, this study investigates both binary classification and multi-class classification. The proposed classification method outperformed the state-of-the-art methods in the literature.