An algorithm for FHR extraction from FHS signals

For foetal surveillance, in the third trimester of pregnancy, the widespread diagnostic tool is ultrasonographic cardiotocography (CTG). However it still presents some limitations. Some anomalies of cardiac functioning are not detectable and, although long-term FHR monitoring is often recommendable, there is no strong evidence how long application of ultrasound energy can be absolutely harmless for foetuses. An alternative can be the phonocardiography, a passive and low cost recording of Foetal Heart Sounds (FHS). A crucial point is the correct recognising of the heart sounds associate to each foetal heart beat for the reconstruction of the FHR signal. Here, an algorithm for FHR extraction from FHS signals is presented. To test the proposed algorithm, about sixty FHS were recorded from pregnant women (35-39 weeks). The measurements were carried out simultaneously with Doppler cardiotocographic recordings for further comparisons and assessments. Audio signals were digitised and stored in a PC. In FHS signals, groups of two bursts can be recognised, which correspond to heart sounds. The first sound (closure of mitral and tricuspid valves) is a good time reference of heart beat, because of its high energy. The proposed algorithm initially employed a cross-correlation with a template to detect the first sound. Subsequently, the non-linear time domain Teager Energy Operator (TEO), designed to enhance areas of local high energy, was employed. By using peaks detection of the envelope of TEO output, it was possible to identify the beats markers. Other steps were necessary to reduce errors due to missed beats, markers due to artefacts, shifted placement of markers. Finally, FHR signal was obtained, with a Reliability Index to express its reliability, and the most common clinical features extracted. Preliminary results seem to be satisfying. In fact, on average, the proposed algorithm provides a percentage of more than 90% of rightly detected beats