Phonocardiography for robust fetal heart rate estimation in real clinical conditions

Fetal heart rate (FHR) analysis is the standard technique for monitoring fetal well-being. Since the non-invasive clinical reference technique for measuring FHR, known as cardiotocography (CTG), suffers from several drawbacks, alternative technologies are now attracting interest. Among them, phonocardiography (PCG) has shown interesting potential. In this study, an original solution is proposed, using a single abdominal PCG (aPCG) to address both the feasibility in clinical routine and the challenge of the difficult detection of temporal events in signals recorded in real clinical conditions. Based on the combination of Non-negative Matrix Factorization (NMF) and Hidden Markov Model (HMM) approaches, the proposed solution exploits the semi-periodicity of the fetal PCG and the temporal continuity a priori of cardiac rhythm to estimate the FHR. Including a pre-processing step of Template subtraction (TS) on aPCG to attenuate maternal component, the proposed algorithm TS+NMF+HMM is evaluated on a database of 38 recordings of pregnant women (total duration of nearly 20 h) and compared to several methods from literature. This new methodological proposition allows a good agreement between estimated FHR and the reference values from CTG (median value of ratio of good agreement at nearly 80% for the whole database) and underlines the interest of using abdominal PCG for fetal well-being monitoring.