Classification of heart sound signals using autoregressive model and Hidden Markov Model

Abstract

This study presents a Computerised Heart Diagnostic System (CHDS) for classifying the different types of heart sounds. A major part of cardiac diagnosis consists of cardiac auscultation. In this study, we developed a simple model, which generates signals for heart sounds. This model could help in identifying the features for assisting in cardiac diagnosis. Additionally, we have also developed a new framework for the CHDS system, which is based on different features of Autoregressive (AR), Mel Frequency Cepstrum Coefficient (MFCC), and Hidden Markov Model (HMM). This system assists in data segmentation, data acquisition, and the time-frequency data transformation, which are generally applied in the AR and the MFCC models in the form of dependable traits. Moreover, this system helps in studying the cardiac auscultation analytically and it helps in monitoring and analysing the complex signals, which represent the heart murmur sounds. Furthermore, the system contains various steps related to data segmentation, signal pre-processing, and steps for pattern recognition. In this study, we have applied many HMM models along with carrying out various experiments for the testing of our model. One major advantage of this system is the fact that it can measure the heart sound signals from several points instead of focusing on a single point. In our study, we have observed that a multi-point heart position provides better and more sensible results as it reflects closely the way the physician examines the subject; listening to the various location before deciding where the targeted sound is best heard. This major issue is not taken into account in most study in this area when designing an automated heart sound analysis system.