Classification of heart sound signal using multiple features and machine learning=Classification des sons cardiaques à l’aide de multiples caractéristiques et de l’apprentissage automatique
No Thumbnail Available
Date
2026
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Université Badji Mokhtar Annaba
Abstract
Cardiovascular diseases (CVDs) are the leading cause of death worldwide, with heart valve disorders-such as aortic and mitral valve anomalies—among the most prevalent. These disorders disrupt the normal mechanical function of the heart, often reflected as acoustic anomalies in heart sounds. Heart sounds, being low-frequency signals generated by the mechanical activity of the heart, carry critical physiological information. Abnormalities such as murmurs or extra sounds (S3, S4) may indicate underlying pathologies. To support early and accessible diagnosis, computer-aided auscultation (CAA) systems have emerged as promising tools, leveraging signal processing and machine learning techniques to classify heart sounds. This thesis presents the development of accurate and efficient methods for automated heart sound classification to assist in the early detection of CVDs. Given the limitations of traditional diagnostic tools— specially in low-resource settings—the study emphasizes the need for affordable and deployable CAA solutions. Several time-frequency representations of phonocardiogram (PCG) signals are explored, including Gammatonegram, Scalogram, Mel-Spectrogram, and Constant-Q Transform (CQT), to identify the most discriminative features. A lightweight architecture based on a novel Directed Acyclic Graph Convolutional Neural Network (DAG-CNN) is proposed and compared with established models such as VGG16, AlexNet, ResNet, and DeiT. Experimental results demonstrate that the proposed approach significantly enhances classification accuracy while reducing computational complexity, making it especially suitable for real-time and resource-constrained environments. This work offers a practical and scalable solution for automated cardiac screening, with potential integration into clinical and remote diagnostic systems.
Description
Keywords
heart sound classification; phonocardiogram; cardiovascular diseases; heart valve disease; computer-aided auscultation; low-frequency signal, time-frequency representation; deep learning; DAG-CNN; signal processing