EPHNOGRAM: A Simultaneous Electrocardiogram and Phonocardiogram Database

Abstract

The electro-phono-cardiogram (EPHNOGRAM) project focused on the development of low-cost and low-power devices for recording simultaneous electrocardiogram (ECG) and phonocardiogram (ECG) data, with auxiliary channels for capturing environmental audio noise, which could be used for PCG quality enhancement through signal processing. The current database, recorded by version 2.1 of the developed hardware, has been acquired from 24 healthy adults aged between 23 and 29 (average: 25.4 ± 1.9 years) in 30min stress-test sessions during resting, walking, running and biking conditions, using indoor fitness center equipment. The dataset also contains several 30s sample records acquired during rest conditions. This data is useful for simultaneous multi-modal analysis of ECG and PCG. It provides interesting insights into the inter-relationship between the mechanical and electrical mechanisms of the heart, under rest and physical activity.

Background

Cardiac auscultation is one of the oldest and most basic methods of cardiac function assessment. Even in the modern cardiac monitoring and cardiac imaging era, the technique remains popular among clinicians, as a preliminary step for screening basic cardiac anomalies. Despite its long history, the visualization, analysis and interpretation of the audio signals acquired from the heart by the phonocardiogram (PCG), is not so common in clinical training. Therefore, PCG-based diagnosis is less common than its electrical counterpart, the electrocardiogram (ECG). Nevertheless, with recent developments in mobile-health and tele-monitoring, the PCG and ECG are again under the spotlight, as low-cost complementary modalities for monitoring the mechanical and electrical functions of the heart [1]. While most research studies in this domain have used separate sessions of ECG and PCG acquisition (which are useful for consistent cardiac anomalies), they do not provide a beat-wise insight into the two cardiac modalities and the inter-relationships between the electro-mechanical functions of the heart.

Methods

The device designed for the EPHNOGRAM project includes circuitry for three-lead ECG, two digital stethoscope channels for PCG acquisition and two auxiliary channels to capture the ambient audio noise. The auxiliary channels are used for digital active noise cancellation (ANC) from the primary PCG channels. The analog signals are filtered by an anti-aliasing analog filter and sampled at 8kHz with a resolution of 12-bits (with 10.5 effective number of bits) and transferred to an on-board low-power microcontroller for minimal preprocessing and registration on a Secure Digital (SD) memory. The device has an LCD and a keypad for basic file naming and controlling the recording duration. Since the quality of the ECG highly depends on the connection of the chest leads, the device provides an online PC-based signal preview feature via Bluetooth, to prevent low-quality signal recordings due to loose body contact.

The front-end anti-aliasing and baseline wander rejection filter consists of a first-order passive high-pass filter with a -3dB cutoff frequency of 0.1Hz, followed by an active 5th order low-pass Butterworth filter, which form bandpass filters that cover the major ECG and PCG bandwidths.

For the ECG front-end, the upper -3dB cutoff frequency was set to 150Hz, with 30dB of attenuation at 1kHz and a 30dB gain in the passband. For the PCG channels, the same active filter topology was used, but with an upper cutoff frequency of 1kHz, 30dB of attenuation at 5kHz, and a passband gain of 5dB.

The ECG and PCG channel filter attenuations were respectively 90dB and 25dB in the Nyquist frequency of the digital front-end ADCs (at 4kHz), which are practically sufficient to avoid aliasing. Note that although the front-end filters are nonlinear phase, the group delays are rather constant over the passbands of the front-end filters. Additional filtering, including power-line cancellation (50Hz for the current database) is performed in the digital domain.

During the prototyping process of the device, various configurations of the hardware and several hand-made stethoscopes were tested. The hardware design version 2.1 was the latest of several progressive improved redesigns, in which the auxiliary channels were added (useful for digital active noise cancellation algorithms) and the signal quality was significantly improved due to a redesign of the analog front-end and the printed circuit board (PCB).

For the stethoscope, the objective was to convert off-the-shelf stethoscopes into high-quality digital ones by using advances signal processing. The different stethoscope designs that were prototyped and tested include: 1) embedding microphones inside standard stethoscope chest-piece (directly under the diaphragm); 2) embedding two microphones inside a standard microphone chest-piece (back-to-back, one facing the diaphragm, the second facing the bell hole); 3) embedding microphones at the end of the tubing (at the junction of the ear tubes); 4) embedding microphones inside the tube, a few centimeters after the stem, plus auxiliary microphones inside the device case to pick environmental audio noises. The latter configuration was found to be the most robust and resulted in high quality signals, with various advantages [2].

The study was approved by the Biomedical Engineering Review Committee (IRB equivalent) of Shiraz University and the individuals gave written informed consent to participate in the study. A total number of 24 male subjects aged between 23 and 29 (average: 25.4±1.9 years) participated in this study.

Data Description

The dataset consists of 69 simultaneous ECG and PCG recordings, each with a duration of 30 seconds (8 records) and 30 minutes (61 records), acquired synchronously from a three-lead ECG and a single PCG stethoscope. Each volunteer performed a specific task once (as detailed in [2]). In a few cases where the data quality was poor (due to electrode/stethoscope detachment and analog front-end saturation), the test was repeated to obtain acceptable data. However, even the poor quality samples have been included in the dataset for noise research purposes and labeled as low quality in the spreadsheet accompanying the dataset. In addition to the main PCG channel, for some subjects the auxiliary audio channels PCG2, AUX1, and AUX2 were recorded for audio processing research purposes; although these auxiliary channels are mostly at quantization noise level for the majority of the recorded sessions.

The 30min records of the dataset (62 records) were acquired in an indoor sports center. A structured interview determined that the participants were in good physical condition and none reported symptoms of autonomic or cardiovascular disorder. Each subject participated in one or a number of the physical scenarios detailed in [2]. Accordingly, ten subjects participated in the first scenario. Five of the participants of the first scenario did not attend the rest of the stages. With the addition of six new volunteers, a total number of eleven subjects contributed in each of the three other scenarios. Only in the bicycle exercise stress-test (Scenario D described in [2]), two volunteers failed the test due to physical fatigue. In order to prepare for each test, volunteers avoided eating food, drinking caffeine, alcoholic drinks and smoking for three hours before the test; but they were permitted to drink water regularly.

Cardiac auscultation is commonly performed from four major chest areas. The Mitral valve (M) sound is heard better at the distal end of the heart, anatomically landmarked between the fifth and sixth ribs on the body surface. The Tricuspid valve (T) sound is heard well on the left side of the heart, between the fourth and sixth ribs. Therefore, we chose the location between the Tricuspid and Mitral landmarks to record the heart sound, to obtain both the first and second heart sounds to a good extent (see [2] for landmarks). During the stress-tests an elastic cotton chest brace was used to fix the stethoscope in place during the experiment.

The data are presented in both MATLAB (ECGPCG00XY.mat) and WFDB (ECGPCG00XY.dat and ECGPCG00XY.hea) formats, with identical base names (XY = 01,..., 69). The MATLAB files are in double-precision floating point format. These files were covered into 16-bit WFDB format by using the mat2wfdb.m function from the WFDB Toolbox. The accuracy of conversion between MATLAB and WFDB formats was assessed per file and per channel, by comparing the signal-to-noise ratio (SNR) of the original MATLAB files vs the WFDB files read by the RDSAMP function of WFDB. All 16-bit WFDB files had an SNR of above 60dB per channel, as compared to the original MATLAB files. Although 60dB is fully acceptable for most applications, researchers seeking double-precision floating point accuracy are advised to use the MATLAB files.

The description of the corresponding physical activities and the unique subject IDs are provided in the spreadsheet: ECGPCGSpreadsheet.csv. A sample MATLAB script named TestHeartRateCalculation.m is provided for basic heart-rate extraction and analysis from the ECG and PCG channels. Additional source codes for analyzing this data are available in the Open-Source Electrophysiological Toolbox (OSET) [3].

Usage Notes

This current dataset is useful for simultaneous multimodal analysis of ECG and PCG, as it provides interesting insights into the inter-relationship between the mechanical and electrical mechanisms of the heart, under rest and physical activity. As proof of concept, in [4], the current dataset was used to compare the heart-rate time-series and the heart-rate variabilities (HRVs) of the subjects during the aforementioned stress-test. It was shown that while the overall trend of the heart-rate time-series obtained from the ECG and PCG are exactly identical (as expected), there are considerable "micro-variations" between them, which reflects the differences between the electrical and mechanical functions of the heart during different levels of physical activity. Specifically, as the heart-rate increases and the RR-intervals of the ECG become shorter, the differences between the R-peaks and the first and second heart sounds (namely S1 and S2) obtained from the PCG channels do not scale at the same rate. There are also notable differences between stochastic features such as the sample entropy of the ECG- and PCG-driven heart-rate time-series.

Another application for the dataset is to use the simultaneous ECG and PCG channels to develop mathematical PCG models for generating synthetic signals. Previous research in electrocardiography, have shown how synthetic ECG generators [5], can be used to develop algorithms for Bayesian filtering and parameter extraction from highly noisy ECG recordings [6]. The current dataset can help researchers to develop similar algorithms for de-noising and automatic parameter extraction from the PCG. Further details and ideas are presented in [2].

As noted in the Data Description, the three channels PCG2, AUX1, and AUX2 (which are available for some of the records), are mostly very weak in amplitude (at quantization noise level). However, through visual inspection and by listening to these audio channels, it is noticed that they have captured some of the electronic device noises and the weak background sounds in the environment. Therefore, although they are not useful for direct utilization, researchers interested in the signal processing aspects of the dataset might find these channels useful for designing adaptive noise cancellers or multichannel blind and semi-blind source separation algorithms.

Due to the type of available chest braces used in this study and the anatomical points selected for better auscultation of the heart sounds, the setup was found to be inappropriate for females during stress tests (multiple attempts in recording signals from female volunteers were unsuccessful). This issue is addressed in future versions of our design, by developing a customized chest brace that would be comfortable for stress tests, while fixing the stethoscope and ECG leads in place.

Any research/publication based on this database is requested to cite the EPHNOGRAM project on PhysioNet, and also references [2] and [3].

Release Notes

Initial release.

Acknowledgements

The database was acquired as part of a master's thesis in Biomedical Engineering at the School of Electrical & Computer Engineering of Shiraz University, Shiraz, Iran between 2016 and 2018 [4].

Conflicts of Interest

The authors declare that there are no conflicts of interest.

References

1. C. Liu, D. Springer, Q. Li, B. Moody, R. A. Juan, F. J. Chorro, F. Castells, J. M. Roig, I. Silva, A. E. W. Johnson, Z. Syed, S. E. Schmidt, C. D. Papadaniil, L. Hadjileontiadis, H. Naseri, A. Moukadem, A. Dieterlen, C. Brandt, H. Tang, M. Samieinasab, M. R. Samieinasab, R. Sameni, R. G. Mark, and G. D. Clifford. An open access database for the evaluation of heart sound algorithms. Physiological Measurement. 2016 Nov 21;37(12):2181. DOI: https://doi.org/10.1088/0967-3334/37/12/2181
2. A. Kazemnejad, P. Gordany, and R. Sameni. An open-access simultaneous electrocardiogram and phonocardiogram database. bioRxiv, 2021. DOI: https://doi.org/10.1101/2021.05.17.444563
3. R. Sameni, The Open-Source Electrophysiological Toolbox (OSET), v 3.14, URL: https://github.com/alphanumericslab/OSET
4. A. Kazemnejad, Analysis of Synchronous Electrocardiogram and Phonocardiogram Parameters Extracted from Normal Subjects, Master's thesis in Biomedical Engineering, School of Electrical & Computer Engineering, Shiraz University, September 2018.
5. P. E. McSharry, G. D. Clifford, L. Tarassenko and L.A. Smith. A dynamical model for generating synthetic electrocardiogram signals. IEEE transactions on biomedical engineering, 50(3), 289-294. 2003. DOI: https://doi.org/10.1109/TBME.2003.808805
6. R. Sameni, M.B. Shamsollahi, C. Jutten and G. D. Clifford. A nonlinear Bayesian filtering framework for ECG denoising. IEEE Transactions on Biomedical Engineering, 54(12), 2172-2185. 2007. DOI: https://doi.org/10.1109/TBME.2007.897817