Pulse Transit Time PPG Dataset

Abstract

We provide an open access, high resolution and time synchronised dataset from multiple sensors worn at different body locations including Photoplethysmogram (PPG), Inertial, Pressure and ECG. The recordings are from 22 healthy subjects performing 3 physical activities. This dataset contains 66 waveform records from multi-site and multi-wavelength PPGs, sensors’ attachment pressures, sensors’ temperatures, inertial data from accelerometer and gyroscope, along with annotated ECG data for a total of 19 channels. Additionally, systolic and diastolic blood pressures, as well as blood oxygenation saturation levels (SpO2) numerics are provided.

Background

Pulse Transit Time (PTT) is the time that pulse waves require to travel in blood vessels between two sites [1]. It is an important indicator for many medical properties [2] and recent research focused on its potential for continuous blood pressure monitors [3, 4]. Three challenges make it difficult to measure PTT precisely: 1) Photoplethysmogram (PPG) measurements are inherently noisy [5, 6], 2) the required wave phase detection accuracy for measurement sited in proximity (e.g., on the same finger) is in the order of milliseconds [7] and 3) the lack of available datasets for PPG measurements in proximity at the required high sample rates. This dataset is designed to address these challenges by providing, amongst other physiologic time series and numerics such as blood pressure, 2x3 unfiltered raw PPG sensor signals of multiple wavelengths for 2 measurement sites in a defined distance to each other.

The dataset contains 66 recordings of 19 physiologic time series for more than 40000 heartbeats. These data were collected to conduct investigations into signal processing and machine learning models for applications in short distance Pulse Transit Time (PTT) recognition, cuffless pressure sensing and other cardiovascular activity modelling research.

Methods

The data were acquired from 22 healthy subjects at The University of Sydney. 6 participants were female and the age range was from 20 to 53 with a mean of 28.52 years. All participants performed 3 activities in random order, namely sitting, stationary walking and running. The data was collected from a device similar to commercial pulse oximeters, containing commercial sensors in a 3D-printed finger clip. A 3-lead ECG recorded the electrical signal of the heart in parallel. The included file device_schematic.png illustrates the used device.

The numerical records include gender, age group, weight group, systolic and diastolic blood pressure at the beginning and end of each activity. The heart rate was recorded at the beginning and end of the measurements with a commercial pulse oximeter and blood pressure monitor. SpO2 levels were measured at the start and end of each activity.

The hardware consisted of the following sensors, their locations is illustrated in the attached file device_schematic.png:

• 2X Maxim Integrated MAX30101 PPG configured to measure
  • infrared λ=880+20−10nm pleth_1 and pleth_4,
  • red λ= 660±10nm pleth_2 and pleth_5,
  • green λ= 537+4−7nm pleth_3 and pleth_6 at 1000Hz (multi-LED mode). The optimum settings for the LED current (18.8mA), pulse width of 215μs and ADC range of 16384 were grid searched
• 2x TAL221100g miniature load cells lc_1, lc_2 measuring the mechanical attachment pressure, connected to 2x HX711 24-bit loadcell amplifier measuring at 80Hz
• 1x TDK - InvenSenseMPU-9250 IMU a_x, a_y, a_z, g_x, g_y, g_z measuring at 500Hz
• 1x AD8232 ECG amplifier ecg measuring at 500Hz

All sensors were connected to an ARM Cortex-M4 at 180 MHz microcontroller, reading all sensors within a 2ms window (500Hz).

The blood pressure was recorded with an OMRON HEM-7322 blood pressure monitor and blood oxygen saturation levels with an iHealth Air Wireless Pulse Oximeter.

The study was approved by the University of Sydney's Human Research Ethics Committee (approval 2020/7059) and all participants gave written informed consent.

Data Description

The data is distributed in two formats, WFDB (WaveForm DataBase) and CSV (comma-separated-value). The WFDB data for all participants have been placed in the root directory along with a corresponding RECORDS file. Each WFDB .hea header file contains the participants' numerics such as <weight>. .atr annotation files contain automatically detected and manually verified R peak ECG annotations for all records. Record, header and annotation file names incorporated subject and activity following this structure:

sXX_YYYY

• s = subject
• XX = subject number (1-22)
• YYY = activity (sit, walk, run)

The data is also provided in CSV format in the \CSV folder. All CSV records include a “time” column that was date shifted to de-indentify participants. The \CSV folder also contains the file subjects_info.csv, describing the participants' numerics such as <weight>.

Both WFDB and CSV records contain the following channels:

• ecg: 3-lead ECG captured at 500Hz
• peaks: CSV ONLY, annotated in WFDB. The annotated ECG R peak (1 = peak, 0 = no peak)
• pleth_1: MAX30101 red wavelength PPG from the distal phalanx (first segment) of the left index finger palmar side (arbitrary units, 500Hz)
• pleth_2: MAX30101 infrared wavelength PPG from the distal phalanx (first segment) of the left index finger palmar side (arbitrary units, 500Hz)
• pleth_3: MAX30101 green wavelength PPG from the distal phalanx (first segment) of the left index finger palmar side (arbitrary units, 500Hz)
• pleth_4: MAX30101 red wavelength PPG from the proximal phalanx (base segment) of the left index finger palmar side (arbitrary units, 500Hz)
• pleth_5: MAX30101 infrared wavelength PPG from the proximal phalanx (base segment) of the left index finger palmar side (arbitrary units, 500Hz)
• pleth_6: MAX30101 green wavelength PPG from the proximal phalanx (base segment) of the left index finger palmar side (arbitrary units, 500Hz)
• lc_1: TAL221 load cell proximal phalanx (first segment) PPG sensor attachment pressure (arbitrary units, 80Hz)
• lc_2: TAL221 load cell (base segment) PPG sensor attachment pressure (arbitrary units, 80Hz)
• temp_1: distal phalanx (first segment) PPG sensor temperature (°C, 10Hz)
• temp_2: proximal phalanx (base segment) PPG sensor temperature in (°C, 10Hz)
• temp_3: InvenSenseMPU-9250 IMU temperature (°C, 500Hz)
• a_x: InvenSenseMPU-9250 IMU acceleration in x-direction (g, 500Hz)
• a_y: InvenSenseMPU-9250 IMU acceleration in y-direction (g, 500Hz)
• a_z: InvenSenseMPU-9250 IMU acceleration in z-direction (g, 500Hz)
• g_x: InvenSenseMPU-9250 IMU angular velocity around x-axis (°/s, 500Hz)
• g_y: InvenSenseMPU-9250 IMU angular velocity around y-axis (°/s, 500Hz)
• g_z: InvenSenseMPU-9250 IMU angular velocity around z-axis (°/s, 500Hz)

Each CSV record includes a time column that was date shifted to de-indentify participants. All WFDB header files or subjects_info.csv contain the following information for each participant:

• <filename>: record filename
• <activity>: sit, walk or run
• <gender>: male or female
• <height>: in increments of 5 (cm)
• <weight>: in increments of 5 (kg)
• <age>: in increments of 5 (years)
• <filename>: record filename
• <activity>: sit, walk or run
• <gender>: male or female
• <height>: in increments of 5 (cm)
• <weight>: in increments of 5 (kg)
• <age>: in increments of 5 (years)
• <bp_sys_start>: systolic blood pressure at the start of the measurement (mmHg)
• <bp_sys_end>: systolic blood pressure at the end of the measurement (mmHg)
• <bp_dia_start>: diastolic blood pressure at the start of the measurement (mmHg)
• <bp_dia_end>: diastolic blood pressure at the end of the measurement (mmHg)
• <hr_1_start>: heart rate as measured with the OMRON HEM-7322 blood pressure monitor at the start of the measurement (bpm)
• <hr_2_start>: heart rate as measured with the iHealth Air Wireless Pulse Oximeter at the start of the measurement (bpm)
• <hr_1_end>: heart rate as measured with the OMRON HEM-7322 blood pressure monitor at the end of the measurement (bpm)
• <hr_2_end>: heart rate as measured with the iHealth Air Wireless Pulse Oximeter at the end of the measurement (bpm)
• <spo2_start>: SpO2 at the start of the measurement (%)
• <spo2_end>: SpO2 at the end of the measurement (%)
• <hr_1_start>: heart rate as measured with the OMRON HEM-7322 blood pressure monitor at the start of the measurement (bpm)
• <hr_2_start>: heart rate as measured with the iHealth Air Wireless Pulse Oximeter at the start of the measurement (bpm)
• <hr_1_end>: heart rate as measured with the OMRON HEM-7322 blood pressure monitor at the end of the measurement (bpm)
• <hr_2_end>: heart rate as measured with the iHealth Air Wireless Pulse Oximeter at the end of the measurement (bpm)
• <spo2_start>: SpO2 at the start of the measurement (%)
• <spo2_end>: SpO2 at the end of the measurement (%)

Usage Notes

The WFDB files can be found in the root directory. The participants are denoted by their filename, e.g. s1_run indicating subject 1, running. There are 3 files for each of the subjects’ 3 activities, for example for subject 1, running:

• s1_run.head is the header file, which also contains participant information such as participant weight
• s1_run.dat contains subject 1 raw recording (19 channels) while running
• s1_run.atr contains the annotations for subject 1’s ECG R peaks

The CSV files containing 19 signals each can be found in the \CSV folder. The ECG R peak annotations are coded in an additional “peaks” column where “1” represents an R peak and “0” no peak. The subjects_info.csv file contains the numerics such as participant weight for all participants.

The data has been used in a study for accurate heart rate monitoring [8]. It can be reused for time series investigations where accurate inter-channel time synchronisation and high time resolution are required, such as pulse transit time derived applications. The data can also be used to investigate pulse wave-induced dynamic pressure changes at fingers.

It should be noted that the sensor data is unfiltered to retain all timing information and to allow direct experimentation, whereas most published PPG data is filtered. To convert the data, we found that removing the DC component by subtracting the output of a centered mean rolling gaussian window and adding a 0.75-5Hz bandpass provided results similar to filtered datasets.

The limitations are that for subject 13 the ECG data is particularly noisy during the walking exercise and that the used InvenSenseMPU-9250 IMU occasionally showed single-sample artifacts, including for temperature.

Release Notes

Version 1.0.0: Initial release

Version 1.1.0: Fixed an issue, <bp_dia_start> and <bp_dia_end> contained duplicates of <bp_sys_start> and <bp_sys_end>

Ethics

The authors declare no ethics concerns.

Acknowledgements

This work was supported by the University of Sydney Cardiovascular Initiative funding. Dr. Withana is the recipient of an Australian Research Council Discovery Early Career Award (DE200100479) funded by the Australian Government.

Conflicts of Interest

No conflicts of interests are declared.

References

1. van Velzen, M.H.N., et al., Increasing accuracy of pulse transit time measurements by automated elimination of distorted photoplethysmography waves. Medical & biological engineering & computing, 2017. 55(11): p. 1989-2000.
2. Foo, J.Y.A. and S.J. Wilson, A computational system to optimise noise rejection in photoplethysmography signals during motion or poor perfusion states. Medical and Biological Engineering and Computing, 2006. 44(1): p. 140-145.
3. Lee, H., H. Chung, and J. Lee, Motion Artifact Cancellation in Wearable Photoplethysmography Using Gyroscope. IEEE Sensors Journal, 2018. PP: p. 1-1.
4. Gesche, H., et al., Continuous blood pressure measurement by using the pulse transit time: Comparison to a cuff-based method. European journal of applied physiology, 2011. 112: p. 309-15.
5. Ghosh, S., et al. Continuous blood pressure prediction from pulse transit time using ECG and PPG signals. in 2016 IEEE Healthcare Innovation Point-Of-Care Technologies Conference (HI-POCT). 2016.
6. Geddes, L.A., et al., Pulse Transit Time as an Indicator of Arterial Blood Pressure. Psychophysiology, 1981. 18(1): p. 71-74.
7. Lane, J.D., et al., Pulse Transit Time and Blood Pressure: An Intensive Analysis. Psychophysiology, 1983. 20(1): p. 45-49.
8. P. Mehrgardt, M. Khushi, S. Poon and A. Withana, Deep Learning Fused Wearable Pressure and PPG Data for Accurate Heart Rate Monitoring, in IEEE Sensors Journal, 2021