Heart Rate Estimation from Phonocardiogram Signals Using Non-negative Matrix Factorization

Electrocardiogram (ECG) is classically considered for heart rate (HR) estimation. However in certain conditions, its use may be difficult and alternative techniques, such as phonocardiograhpy (PCG), are investigated. For PCG signals, in most studies, the challenge is to detect and annotate the heart sounds ${S_1}$ and ${S_2}$, which may become quasi-impossible in case of noise. In this paper, we present a novel approach of HR estimation from PCG signals based on non-negative matrix factorization (NMF), applied to the spectrogram of PCG, considered as a source-filter model. Compared to state of the art methods, specific considerations based on the signal properties have been included to ensure the reliability of the decomposition. HR estimations obtained from noise-free and noisy real PCG signals are evaluated by comparison to HR estimation from synchronous ECG.