PCG-Delineator: An Efficient Algorithm for Automatic Heart Sounds Detection in Fetal Phonocardiography

Abstract

Fetal phonocardiography (FPCG) is a non-invasive electronic recording of the acoustic cardiac signals. Unfortunately, FPCG is hidden by high-amplitude noise which makes detection of FPCG waveforms challenging. Aim of the study is to propose PCG-Delineator as an algorithm for automatic detection of the first and second heart sound (S1 and S2, respectively) from FPCG. To this aim, 37 simulated FPCG tracings (Physionet) are filtered by a wavelet-based procedure (4th order Coiflets mother wavelet with 7 decomposition levels) to erase noise. Successively, S1 and S2 are detected. S1 detection procedure is threshold-based (threshold=30% of the filtered FPCG signal maximum amplitude), under the condition that 40ms separate two consecutive S1 sounds. S2 detection procedure is also threshold-based, but under the conditions that S2 has to fall 100ms after preceding S1 and 200ms before successive S1, and that S2 has to have an amplitude lower than 80% that of preceding S1. Sensitivity (SE) and positive predictive values (PPV) were computed. Results indicate that PCG-Delineator was able to reduce noise (our SNR: from -1.1÷7.4dB to 12.9÷17.9dB; P<10-14) and to accurately detect both S1 (SE: 88%; PPV: 91%) and S2 (SE: 77%; PPV:99%). In conclusion, PCG-Delineator is an efficient algorithm for automatic heart sounds detection in FPCG.