Heart Sound Analysis Based on Extended Features and Related Factors

The heart sound classification aims to explore a reasonable method to distinguish normal heart sounds from abnormal heart sounds. In the traditional heart sound classification, their performance is limited because of the baseline feature of heart sound. This paper investigate the factors that affect the performance of classification tasks and extend the corresponding features. In this work, several basic features and extened features of the heart sound are evaluated based on support vector machine (SVM). In the initial stage, a phonocardiogram(PCG) signal is divided into several segments and 64 features in total should be extracted from time domain and frequency domain, these features are divided into 8 groups according to different property. Feature extension is made for the corresponding results, the frequency domain characteristics of diastole and systole are extended as new features, all these new features are added to the original features and compared in different classification method. Finally, an ensemble learning method is proposed, results for the normal and abnormal heart sound signal, the sensitivity, specificity and over all score are 95.9%, 91.7% and 93.8% respectively.