A novel method based on a combination of discrete wavelet transforms, group-based Sparse and tensor decomposition for Heart Sound Classification

Listening to the sound signals of the heart is considered one of the most non-invasive and easy ways to diagnose the irregularities of the human heart, the correct analysis of which requires the knowledge and experience of a specialist doctor. The purpose of this paper is to design and implement a computer-aided diagnosis (CAD) system for detecting and classifying normal and abnormal heart sounds from phonocardiogram (PCG) signals. To perform experiments, the PhysioNet database was used. In the pre-processing step, noise and environmental disturbances in the PCG signals are removed using band-pass Butterworth filters. Then, discrete wavelet transforms (DWT), group-based Sparse, and tensor decomposition are used to extract features from PCG signals. Finally, the support vector machine (SVM), the k-nearest neighbors (KNN), naive Bayes (NB), the classification and regression tree (CART), and multi-layer perceptron (MLP) were used for the classification step. The employment of DWT, group-based sparse, and tensor decomposition for detection features is the novelty of this paper. The proposed method demonstrated better performance compared to other methods used in different papers. The proposed DWT, group-based sparse and tensor decomposition-NB method had a high accuracy rate of ۹۵.۳%. Also, the combination of PCG feature extraction methods increases the accuracy of the CAD system in diagnosing abnormal heart sounds. The proposed method in this paper uses different methods for extracting features, and their classification has high accuracy for abnormal sound detection.