SensSmartTech database of cardiovascular signals synchronously recorded by an electrocardiograph, phonocardiograph, photoplethysmograph and accelerometer

Abstract

The SensSmartTech database comprises cardiovascular signals synchronously recorded by electrocardiograph (ECG), phonocardiograph (PCG), photoplethysmograph (PPG) and accelerometer (ACC) at heart rates at rest and after activity (HRs). It is composed of 338 30-s recordings taken from 32 healthy volunteers. Recordings contain information from 10 channels: 4 ECG (limb, V3 and V4 leads), 1 PCG (measured at heart apex), 4 PPG (measured at carotid and brachial arteries at two light wavelengths each) and 1 ACC (measured between V3 and V4 ECG electrodes), along with the recording time of each signal. The original recordings are stored in CSV format. Resampled recordings are stored in WFDB format. The large HR range with the average minimum on the set of 83 ± 11 bpm and the average maximum on the set of 143 ± 14 bpm, along with several recordings per heart relaxation period, enable studying of the cardiovascular dynamics and gaining insights into the corresponding cardio-respiratory and electro-mechanical couplings. Of particular clinical significance are the opportunities to derive corrections of multiparameter biomarkers for HR.

Background

Noninvasive multiparametric assessment of cardiovascular function is emerging as a cost-effective method for diagnosing heart failure and arterial diseases, and for telemedical monitoring of blood pressure and neural disorders. This approach involves the simultaneous acquisition of electrocardiographic, phonocardiographic, arterial-pulse, chest-vibration, bioimpedance, and other waveforms [1-4]. Several multisensory databases capturing variations in HR during activity have been documented [5-8]. SensSmartTech stands out as the first base of multisensory recordings which systematically follows heart relaxation dynamics across a wide range of HRs (58 bpm - 173 bpm). The recorded HR-dependence is of interest to clinicians applying the HR biomarker correction, engineers investigating HR estimation by different wearable sensors and the impact of noises and artefacts on diagnostic signals, and scientist studying the underlying nonlinear dynamics of the heart as an electro-mechanical system.

Methods

Data collection was conducted using a custom-made multisensory acquisition device – a polycardiograph, designed with a modular structure to capture and synchronize PCG, ECG, PPG and ACC signals as follows:

• PCG signal is captured using a microphone ICS-40300 (TDK InvenSense) placed in a cardiology stethoscope SPIRIT CK-S474SPF63 (Spirit Medical) with the sampling rate of 1 kHz. 1 PCG stethoscope was positioned at the sternum to the right of V3 ECG electrode and secured with an elastic band.
• ECG signal acquisition is performed with the ADS1298 chip (Texas Instruments) with the sampling rate set to 500 Hz. Measurement used 4 limb electrodes, V3 and V4, while the redundant precordial electrodes (V1, V2, V5 and V6) were placed on the upper right arm to prevent noise from the hanging leads.
• PPG sensors were based on the MAX86150EFF+T chip (Maxim Integrated) with a two-colour oxymeter (660nm and 800 nm) and the sampling rate set to 100 Hz. PPG sensors were attached with inelastic Velcro tapes; one placed around the neck over the left carotid, the other around the upper left arm over the brachial artery.
• ACC signal was recorded by a MEMS accelerometer MPU6050 (TDK InvenSense) with an acceleration range set to +/- 1g. It was attached to the body between V3 and V4 ECG electrodes using a self-adhesive ECG electrode. Only the z axis in the direction perpendicular to the chest was used.

Sensor output signals were digitalized by 16-bit A/D converters. The polycardiograph synchronously collected data from the sensors and transmitted them to a PC over Ethernet. Accuracy of the polycardiograph was set by the sampling rates of the sensors.

The subjects were over 18 years old and without a history of cardiovascular conditions or diseases including myocardial infarction, heart failure, stenosis, dangerous arrhythmias or prescription. Volunteers with benign arrhythmias were included in the study.  

Recordings were taken in a standing position at rest and immediately after the activity. After each recording, the researcher calculated the heart rate (HR). Three 30-second recordings were made at rest. After the activity, recordings were repeated until the HR dropped to 10-20 bpm above the HR at rest.

Data Description

The dataset comprises 338 30-sec polycardiographic signals, each containing 10 channels: 4 ECG, 4 PPG, 1 PCG, and 1 ACC channel. These signals originate from 32 subjects (18 females and 14 males) with an average age of 34.0 ± 8.6 years and an average body-mass index of 25.1 ± 4.3 kg/m².

The dataset is distributed in two formats: WFDB (WaveForm DataBase) and CSV (comma-separated-value) format, in the folders with the corresponding names. CSV files contain original signals. WFDB files contain signals resampled to 1kHz and share a common time (sample) axis.

Separate files are provided for each recording and measurement modality. The files are identified by the subject number (##) and the start time of recording in the format hh-mm-ss.

WFDB files:

• ##_time_ecg.dat contains 4 columns corresponding to ECG lead I, lead II, lead V3, and lead V4,
• ##_time_ppg.dat contains 4 columns corresponding to the carotid pulse wave measured using 660 nm light, the carotid pulse wave measured using 800 nm light, the brachial artery pulse wave measured using 660 nm light, and the brachial artery pulse wave measured using 800 nm light, respectively
• ##_time_acc.dat contains a column corresponding to the accelerometer signal
• ##_time_pcg.dat contains a column corresponding to the phonocardiographic signal

CSV files are named and structured in the same way. However, they contain an additional column (set at the first) providing the information on the real acquisition time of each sensor. The acquisition time follows the sampling rate of the sensor. Sensors may record signals at different points in time. Therefore, the time axes of different sensors are different, but the acquisition is synchronized so that they can be extended to a common zero.

Additionally, a table Demographics.csv lists file names and subject demographics, including age, height, weight, and body-mass index. Furthermore, each row in this table displays the subject activity status: 'B' for the measurement before and 'A' for the measurement after the activity, and the HR calculated as the inverse of the median RR interval per recording.

To de-identify the data, all dates were removed from the recordings. The published data do not contain any information that identifies or provides a reasonable basis to identify an individual. The data comply with HIPPA requirements for sharing personal health information.

Usage Notes

The ECG signals are filtered by a low-pass filter with a 150 Hz cut-off using butter and sosfiltfilt Python functions from the scipy.signal library. The line interference at 50 Hz and its 2^nd and 3^rd harmonics are eliminated using zero-phase Python notch filters iirnotch and filtfilt from the scipy.signal library. The mean value of each PPG signal was set to zero. Considering the valuable information contained in the baseline wander, including the respiration patterns, the data are otherwise left unprocessed and are presented as originally recorded.   

The data can be used to investigate the heart rate estimation by different sensors and the dependence of the QT interval and heart rate variability on heart rate. However, the main advantage of SensSmartTech database is the possibility of combining signals from different sensors to estimate electro-mechanical characteristics of the cardiovascular system, such as systolic time intervals and transient pulse times. The former is used in diagnostics of heart failure and the latter in cuff-less blood pressure monitoring. Given that the early symptoms of heart failure manifest during the exercise, the dataset obtained during relaxation after the activity may provide valuable information on cardiac and arterial limitations at the onset of this condition.

Release Notes

Version 1.0.0: Initial release

Ethics

The study received approval from the Ethics Committee for Work with Human Material at INN Vinca (registry numbers 116-18-2/2022-000, 24-DMS0-029060).

Acknowledgements

The research was supported by the Science Fund of the Republic of Serbia, Grant. No. 7754338, Multi-SENSor SysteM and ARTificial intelligence in service of heart failure diagnosis – SensSmart and the Ministry of Science RS, Grants No. 451-03-47/2023-01/200017 and 451-03-47/2023-01/200103. The authors thank the volunteers who supported the study and A. Maluckov, M. Miletić and U. Ralević for their valuable advice and technical support.

Conflicts of Interest

The authors declare no conflicts of interest.

References

1. Dehkordi P, Khosrow-Khavar F, Di Rienzo M, Inan OT, Schmidt SE, Blaber AP, Sørensen K, Struijk JJ, Zakeri V, Lombardi, Shandhi MMH, Borairi M, Zanetti JM, Tavakolian K (2019). “Comparison of Different Methods for Estimating Cardiac Timings: A Comprehensive Multimodal Echocardiography Investigation”. Front. Physiol. 10, 1057. PMID: 31507437. [Online]. Available from: https://doi.org/10.3389/fphys.2019.01057
2. Weissler AM, Harris WS, Schoenfeld CD (1968). “Systolic Time Intervals in Heart Failure in Man.” Circulation 37, 149–159. PMID: 5640345. [Online]. Available from: https://doi.org/10.1161/01.CIR.37.2.149
3. Ji L, Li P, Liu C, Wang X, Yang J, Liu C (2016). “Measuring Electromechanical Coupling in Patients with Coronary Artery Disease and Healthy Subjects.” Entropy 18, 153. [Online]. Available from: https://doi.org/10.3390/e18040153
4. Elgendi M, Fletcher R, Liang Y, Howard, N, Lovell, NH, Abbott, D, Lim K, Ward R (2019). “The use of photoplethysmography for assessing hypertension.” npj Digital Medicine 2:60. [Online]. Available from: https://doi.org/10.1038/s41746-019-0136-7
5. Jarchi D, Casson, AJ (2016). “Description of a Database Containing Wrist PPG Signals Recorded during Physical Exercise with Both Accelerometer and Gyroscope Measures of Motion.” Data. 2(1), 1. [Online]. Available from: https://doi.org/10.3390/data2010001, [Online]. Available from: https://doi.org/10.13026/C2PQ1X
6. Areias Saraiva J, Abreu M, Carmo A S, Plácido da Silva H, Fred A. ScientISST MOVE: Annotated Wearable Multimodal Biosignals recorded during Everyday Life Activities in Naturalistic Environments (version 1.0.0). PhysioNet. 2023. [Online]. Available from: https://doi.org/10.13026/sg89-qq52
7. Mehrgardt P, Khushi M, Poon S, Withana A. Pulse Transit Time PPG Dataset (version 1.1.0). PhysioNet. 2022. [Online]. Available from: https://doi.org/10.13026/jpan-6n92
8. Goldberger A, Amaral L, Glass L, Hausdorff J, Ivanov PC, Mark R, Mietus JE, Moody GB, Peng CK, Stanley HE (2000). “PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals.” Circulation 101 (23), e215–e220. PMID: 10851218, [Online]. Available from: https://doi.org/10.1161/01.CIR.101.23.e215