CardioWave-Portable: a portable ECG and PPG dataset for cuffless blood pressure estimation

@article{PhysioNet-cardiowave-portable-1.0.0,
  author = {Yang, Hailin and Zhang, Lexi and Liu, Hongyu and Wang, Chao and Xie, Zhenzhen and Guo, Xitong},
  title = {{CardioWave-Portable: a portable ECG and PPG dataset for cuffless blood pressure estimation}},
  journal = {{PhysioNet}},
  year = {2026},
  month = sep,
  note = {Version 1.0.0},
  doi = {10.13026/yasp-vq73},
  url = {https://doi.org/10.13026/yasp-vq73}
}

@article{pollard_physionet_2026, title = {{PhysioNet} as a global platform for biomedical research}, volume = {1}, issn = {3005-0693}, url = {https://doi.org/10.1038/s44360-026-00096-z}, doi = {10.1038/s44360-026-00096-z}, abstract = {Twenty-five years after its first public repository was established, the PhysioNet resource for biomedical and clinical data has enabled tens of thousands of research papers. Further efforts are ongoing to advance secure data sharing, community engagement and scalable research infrastructure.}, number = {8}, journal = {Nature Health}, author = {Pollard, Tom and Moody, Benjamin E. and Lehman, Li-wei H. and Gow, Brian J. and Fernandes, Chrystinne and Xie, Chen and Johnson, Alistair and Mark, Roger G. and Heldt, Thomas}, month = aug, year = {2026}, pages = {792--795},}

Abstract

CardioWave-Portable is a controlled-access dataset of synchronized single-lead electrocardiography (ECG) and photoplethysmography (PPG) signals acquired with a portable card-type device and paired with cuff-based systolic and diastolic blood pressure measurements. The release contains 3,849 fixed-length 24-second waveform records from 977 adult participants, together with record-level blood-pressure labels, signal-quality annotations, and questionnaire-derived participant metadata.

The waveforms are provided as standard WFDB records, with record-level and participant-level metadata provided as CSV files. The dataset is intended to support research on cuffless blood-pressure estimation, physiological signal processing, signal-quality-aware modelling, participant-level evaluation, and reproducible benchmarking. It is a research dataset and does not establish clinical equivalence to validated cuff-based devices.

Because the dataset contains human physiological signals and health-related indirect identifiers, the complete data files are intended for credentialed controlled access under the applicable PhysioNet data-use agreement.

Background

Reliable assessment of cuffless blood-pressure methods requires physiological waveform datasets with clearly documented acquisition conditions, reference measurement procedures, quality information, participant grouping, and reusable file formats. Photoplethysmography captures peripheral pulse-wave dynamics, while electrocardiography provides a timing reference for cardiac electrical activation [1-4].

CardioWave-Portable was created to support waveform analysis, quality-aware modelling, metadata-assisted prediction, participant-level evaluation, and carefully designed calibration studies using signals acquired with a portable device. Current evidence emphasizes important methodological and clinical limitations in cuffless blood-pressure estimation, including calibration dependence, evaluation design, population generalizability, and the distinction between technical prediction performance and clinical validation [5,6]. CardioWave-Portable is therefore provided as a research resource and does not establish clinical equivalence to validated cuff-based devices.

Because the dataset contains physiological waveforms and health-related indirect identifiers, the complete files are intended for controlled access rather than anonymous open download.

Methods

Study design, site, and recruitment

Data were collected during 2026 at a tertiary care hospital in Nenjiang, Heilongjiang Province, China. Adults were recruited through community-screening and hospital-based recruitment at the study site. The cohort included adults without a reported hypertension diagnosis, adults with diagnosed hypertension, and adults reporting blood-pressure-related medication use. Participants provided informed consent, completed a questionnaire, and underwent the standardized acquisition protocol.

The study was reviewed and approved by the Ethics Review Committee of the School of Economics and Management, Harbin Institute of Technology, approval number 2026-12, approved on 7 April 2026. All participants retained in the release were at least 18 years old.

Sample size and demographics

The final controlled-access release contains 3,849 records from 977 adult participants. At participant level, 664 were coded female, 302 were coded male, and sex was unavailable for 11. Mean age was 48.27 years, with a standard deviation of 15.79 years and a range of 18 to 90 years or older. Ages of 90 years or older are top-coded as 90+ in the released participant metadata.

Height was available for 961 participants, with a mean of 164.13 cm, standard deviation of 8.11 cm, and range of 130 to 195 cm. Weight was available for 955 participants, with a mean of 66.56 kg, standard deviation of 13.60 kg, and range of 36 to 168 kg. Diagnosed hypertension was recorded for 265 participants, no diagnosis for 694, and missing for 18. Blood-pressure-related medication use was recorded for 224 participants, no use for 739, and missing for 14.

At record level, mean cuff-based systolic blood pressure was 119.86 mmHg, with a standard deviation of 21.28 mmHg and a range of 68 to 210 mmHg. Mean cuff-based diastolic blood pressure was 73.22 mmHg, with a standard deviation of 13.10 mmHg and a range of 40 to 140 mmHg.

Equipment and acquisition protocol

ECG and PPG were acquired using a ChoiceMed MD100S portable card-type device. The device produced one single-lead ECG channel and one PPG channel sampled at 500 Hz. A raw acquisition lasted approximately 30 seconds and contained approximately 15,000 samples per channel. Two coated hardware versions within the same device family were used: blue, coded b, and gold, coded g. Each participant was scheduled for two blue-card and two gold-card measurements.

Before acquisition, participants sat upright and rested for approximately 10 minutes. The arm was maintained approximately at heart level. Reference blood pressure was measured on the left upper arm with an OMRON OML-HEM-1026 cuff-based monitor immediately after each card-device acquisition. Cuff inflation and deflation therefore did not occur during the released ECG and PPG segment.

Waveform preparation

Samples with zero-based indices 1,000 through 12,999 were retained from each raw recording. This corresponds to the half-open interval from 1,000 inclusive to 13,000 exclusive. Each released signal contains 12,000 samples, equivalent to 24 seconds at 500 Hz.

The stored waveform values are raw device digital samples. No physical amplitude calibration, resampling, baseline-removal filter, band-pass filter, or low-pass filter was applied to the stored waveform values. Manufacturer conversion factors to calibrated physical amplitudes were unavailable. The WFDB headers therefore provide a documented normalized display view using normalized units, fixed channel-level gains, and record-specific and channel-specific median baselines. This header display metadata does not alter the stored 32-bit digital samples.

Signal-quality annotation

Each raw recording was manually reviewed separately for ECG and PPG in three overlapping windows: samples 0 through 4,999; samples 4,000 through 8,999; and samples 8,000 through 12,999. Each window was labelled acceptable or poor according to waveform completeness, rhythmic regularity, baseline drift, severe noise, and usability. The released ECG and PPG quality labels equal the number of poor windows and range from 0 to 3. Records were not excluded based on these quality labels.

Dataset construction and quality control

The internal source contained 4,228 raw recordings from 1,057 participants. File and label matching and basic metadata checks retained 3,915 records from 980 participants. Duplicate-source removal excluded 58 records, leaving 3,857 records from 979 participants. During resubmission quality control, eight records from two participants younger than 18 were removed. Five implausible height values were corrected using retained questionnaire-entry and review material. Sex and age values that differed across repeated records were harmonized to the unique non-missing value in the corresponding same-session source entry or review material. The final release contains 3,849 records from 977 adult participants.

The complete transformation rules, exclusion criteria, quality-control procedures, and plausibility checks are documented in CLEANING_AND_FILTERING.md.

Temporal privacy

Actual record-level acquisition dates and times and internal temporal labels are not included in the release. The release excludes record-level dates, times, and years, dates of birth, original filenames, study-staff identifiers, and original participant identifiers. WFDB headers contain no base date or base time. The year 2026 describes only the overall study period and is not linked to individual records.

Data Description

Waveforms are provided in WFDB format. Each record consists of one .hea header and one .dat binary signal file containing synchronized ECG and PPG. Each record contains two signed 32-bit digital channels sampled at 500 Hz for 12,000 samples, corresponding to 24 seconds.

For display, the WFDB headers use normalized units. ECG uses a fixed gain of 1,000 digital counts per normalized unit, and PPG uses a fixed gain of 12,000 digital counts per normalized unit. Each record and channel uses the rounded digital median as its display baseline. The CALIBRATION file supplies recommended plotting scales. These header values affect only the normalized display view; the signed 32-bit values stored in the .dat files are unchanged.

The file metadata/records.csv links each waveform record to its participant, cuff-based systolic and diastolic blood pressure, card type, fixed source segment indices, and ECG and PPG quality labels. The top-level subject-info.csv file contains one row per participant and includes questionnaire-derived demographic, anthropometric, hypertension-related, medication, lifestyle, and pre-measurement variables. DATA_DICTIONARY.csv defines each field, data type, unit or code, missing-value representation, and description. RECORDS lists all WFDB record paths without file extensions.

Waveforms are divided into 16 numbered upload shards named part-0001 through part-0016. Each of the first 15 directories contains no more than 500 files, and part-0016 contains the remaining records. Sharding is organizational only. Participant membership is provided by subject_id in metadata/records.csv.

Additional files include README.md, CLEANING_AND_FILTERING.md, PRIVACY.md, ACCESS.md, REFERENCES.md, an example WFDB loading script in the examples directory, and validation reports.

Usage Notes

Users should read README.md, DATA_DICTIONARY.csv, and CLEANING_AND_FILTERING.md before using the dataset. The complete waveform and metadata files are provided under controlled access because they contain human physiological signals and health-related indirect identifiers.

Repeated records from the same participant must remain in the same data split. Random record-level splitting could place records from one participant in both training and evaluation sets and produce overly optimistic estimates because of participant leakage. Users should report any additional filtering, normalization, missing-data handling, cohort restrictions, calibration procedure, and participant-level data-splitting method.

The normalized waveform amplitudes are not calibrated physical amplitudes. Users who require the exact stored device values should read the WFDB records in digital mode. Quality labels are manual annotations and were not used to exclude records from the complete release.

Credentialed users must comply with the applicable PhysioNet data-use agreement. Users must not attempt to identify or contact participants, link the data to external personal information, redistribute controlled files to unauthorized persons, or otherwise use the data outside the approved conditions. Publications and derived works should cite the associated dataset and Data Descriptor when available.

Release Notes

Version 1.0.0

Initial Release

Ethics

The data collection and study protocol were approved by the Ethics Committee / Institutional Review Board of the School of Management, Harbin Institute of Technology (Approval No. 2026-12). All participants provided written informed consent before participation. The consent process covered participation in the study and the use of de-identified data for research purposes.

The released database does not include direct personal identifiers such as names, contact information, addresses, identity numbers, or facial images. The shared data files contain de-identified physiological waveform records, cuff-based blood pressure labels, demographic and anthropometric variables, hypertension-related diagnosis and medication-use status, and pre-measurement lifestyle/status variables. Because these data include human physiological waveforms, multiple indirect identifiers, and health-related sensitive variables, the full database is shared under credentialed controlled access with a Data Use Agreement rather than anonymous open download.

Users granted access are required to comply with the applicable Data Use Agreement, including restrictions against participant re-identification, redistribution to unauthorized users, attempts to contact participants, and insecure data handling.

Acknowledgements

This work was supported by the National Natural Science Foundation of China (Nos. 72504069, 72125001, 72431004, 72071054, 72121001, and 72293584). The authors thank Professor Zhenzhen Xie for providing the experimental environment and resources that made this study possible.

Conflicts of Interest

The authors declare no competing interests.

References

1. Goldberger AL, Amaral LAN, Glass L, et al. PhysioBank, PhysioToolkit, and PhysioNet: Components of a new research resource for complex physiologic signals. Circulation. 2000;101(23):e215-e220. doi:10.1161/01.CIR.101.23.e215.
2. Allen J. Photoplethysmography and its application in clinical physiological measurement. Physiological Measurement. 2007;28(3):R1-R39. doi:10.1088/0967-3334/28/3/R01.
3. Elgendi M, Fletcher R, Liang Y, et al. The use of photoplethysmography for assessing hypertension. NPJ Digital Medicine. 2019;2:60. doi:10.1038/s41746-019-0136-7.
4. Mukkamala R, Hahn JO, Inan OT, et al. Toward ubiquitous blood pressure monitoring via pulse transit time: Theory and practice. IEEE Transactions on Biomedical Engineering. 2015;62(8):1879-1901. doi:10.1109/TBME.2015.2441951.
5. Mehta S, Kwatra N, Jain M, McDuff D. Examining the challenges of blood pressure estimation via photoplethysmogram. Scientific Reports. 2024;14(1):18318. doi:10.1038/s41598-024-68862-1.
6. Yang E, Schutte AE, Stergiou G, et al. Cuffless blood pressure measurement devices: International perspectives on accuracy and clinical use. JAMA Cardiology. 2025;10(6):624-631. doi:10.1001/jamacardio.2025.0662.