Signal Statistics of Heart Sound Recordings: A Comparative Study Between Smartphones and Electronic Stethoscopes

Smartphones possess the potential to become widely available stethoscopes for cardiac diagnosis due to their audio processing capabilities and widespread usage. However, the quality of signals collected by phones should be characterised and compared to the signals collected by clinical stethoscopes. To this ends, this study compares the characteristics of cardiac auscultation signals captured by either smartphones or electronic clinical stethoscopes on ten subjects. Results revealed that the electronic stethoscope captures sharper and cleaner signals than the smartphone (kurtosis of 29.381 ± 29.436 vs. 4.658 ± 4.301, P < 0.001), as well as a wider frequency response range (median frequency 64.60 ± 16.65 Hz vs. 49.96 ± 11.05 Hz, P = 0.044). Besides, in the key frequency bands 1-3
^rd
of the Mel-frequency cepstral coefficients (MFCCs), the correlation coefficients show a diverse behaviour, with the 1
^st
and 3
^rd
being high (average 0.993 and 0.896), while the 2
^nd
decreases to 0.428. This implies that the time-frequency domain features extracted from the two signal groups differ, and directly transferring a model trained on electronic stethoscope data to smartphones may result in suboptimal performance. These findings illustrate the clear distinctions between using smartphones and electronic stethoscopes for auscultation. The characteristics provide valuable insights for signal quality metrics to enhance the accuracy in identifying cardiac issues and the feasibility of remote auscultation.