Simulated Fetal Phonocardiography Signals (PPKE-ITK)

Simulation details

Generator software based on Cesarelli et al. [1]. First an S1-S1 interval sequence was generated based on a power spectral density modeled by a bimodal Gaussian distribution with means: 0.1, 0.25, standard deviations: 0.01, 0.03. The relative powers were: 0.15, 0.03. This was converted to a heart rate signal, which then determined the locations of the S1 sounds. Other parameters specific for the heart sounds (gestational week-GW, mean fetal heart rate-FHR, standard deviation of fetal heart rate-SD) or noises were not changed from the original.

Parameter combinations (normal case)

• GW: 34, FHR: 140, SD: 2
• GW: 34, FHR: 135, SD: 3
• GW: 36, FHR: 135, SD: 3
• GW: 36, FHR: 130, SD: 2
• GW: 38, FHR: 130, SD: 4
• GW: 38, FHR: 130, SD: 2

Noise levels (normal case)

• no maternal heart sound, no limited duration impulses, SNR: approx. -4.1 dB
• no limited duration impulses, SNR: approx. -4.3 dB
• no limited duration impulses, SNR: approx. -8.1 dB

Parameter combinations (noisy case)

• GW: 34, FHR: 140, SD: 2
• GW: 36, FHR: 135, SD: 3
• GW: 38, FHR: 130, SD: 4

Noise levels (noisy case)

• -4.3 dB
• -6.6 dB
• -8.1 dB
• -10.2 dB
• -11.2 dB
• -14.7 dB
• -15.3 dB
• -16.4 dB
• -17.7 dB
• -19.1 dB

Files

Sound files

Simulated fetal phonocardiograms are stored as mono 16-bit PCM wave files, with 1000 Hz sampling frequency.

Labels

The individual heart sound locations are stored in seconds and labeled as "S1" or "S2". These are stored in a csv format with a delimiter of a ';'. A header is included marking the "Location" and "Value" columns. These files are stored alongside the sound files.

Filename structure

All relevant information can be extracted from the filename of the given record:

GW<gestational_week>_<mean fetal heart rate>_<standard deviation of fetal heart rate>_SNR<signal to noise ratio in dB>dB_<"noLDI" if there are no limited duration impulses>_m<mean maternal heart rate>_<index>

Notes

For all parameter combinations and noise levels multiple signals were generated. All signals can be considered independent since the random generator was seeded only once.

For more details about the generator and the other parameters, see: [1]

References

1. Cesarelli M, Ruffo M, Romano M, Bifulco P. Simulation of foetal phonocardiographic recordings for testing of FHR extraction algorithms. Comput Methods Programs Biomed 2012 Sep;107(3):513-23.