The Hb-PPG Dataset: A Milestone in Optical Biomedical Sensing
Researchers have unveiled a groundbreaking dataset that could revolutionize non-invasive hemoglobin monitoring. The Hb-PPG dataset, published today in Nature Scientific Data, provides high-quality multi-wavelength photoplethysmography (PPG) signals paired with clinical reference measurements.
Led by scientists from Guilin University of Electronic Technology in China, with key contributions from Mohamed Elgendi at Khalifa University in the UAE, the Hb-PPG resource addresses critical gaps in existing data for machine learning models. Elgendi's involvement underscores Khalifa University's growing prominence in AI-driven biomedical engineering.
Understanding PPG and Its Role in Health Monitoring
PPG technology powers many consumer wearables, measuring pulse rate and oxygen saturation by analyzing light absorption variations caused by arterial blood flow. For hemoglobin—a protein in red blood cells essential for oxygen transport—traditional invasive methods like venipuncture remain the gold standard but are painful, costly, and unsuitable for continuous monitoring.
Non-invasive alternatives using PPG aim to estimate hemoglobin levels (Hb) by leveraging light wavelengths where absorption differs between oxygenated and deoxygenated blood, as well as hemoglobin concentration. The Hb-PPG dataset uses four specific wavelengths: 660 nm (red), 730 nm (near-infrared edge), 850 nm, and 940 nm (infrared), chosen to minimize interference from skin pigmentation, motion, and other artifacts.
This multi-wavelength approach enhances model robustness, a common challenge in single-wavelength systems like those in early smartwatches.
Dataset Details: Scale, Diversity, and Accessibility
The Hb-PPG dataset includes 1008 PPG signal segments from 252 adults aged 21 to 90 years. Each segment captures simultaneous signals at the four wavelengths, synchronized with reference lab measurements of hemoglobin (via standard blood tests), fasting blood glucose, systolic blood pressure, and diastolic blood pressure. Subjects represent a diverse range, aiding in generalizable AI models.
- Subjects: 252 adults (balanced age distribution 21-90)
- Signals: 1008 segments, 60 seconds each, sampled at 125 Hz
- Reference Data: Hb (g/dL), glucose (mmol/L), SBP/DBP (mmHg)
- License: Creative Commons CC BY 4.0
Freely available on Figshare (download here), with validation code on GitHub. This openness fosters global collaboration, especially valuable for resource-limited settings combating anemia.
Khalifa University's Pivotal Contribution Through Mohamed Elgendi
Mohamed Elgendi, Assistant Professor in Khalifa University's Department of Biomedical Engineering and Biotechnology, played a crucial role in signal processing, validation, and clinical integration. Funded partly by Khalifa University grant FSU-2025-001, his expertise in PPG analysis—evidenced by over 170 publications and top 2% global scientist ranking—elevates the dataset's credibility.
Elgendi's prior work on PPG for blood pressure and cardiovascular monitoring aligns perfectly, positioning Khalifa University as a leader in UAE's biomedical innovation ecosystem. For aspiring researchers, opportunities abound at higher-ed-jobs in UAE universities like Khalifa.
The Global Anemia Crisis: Why Accurate Monitoring Matters
Anemia affects nearly 2 billion people worldwide, with WHO estimating 40% of children 6-59 months, 37% pregnant women, and 30% women 15-49 years impacted.
Real-time, non-invasive screening via wearables could transform detection in remote UAE areas or during routine checkups, reducing reliance on labs.
Technical Challenges in PPG-Based Hb Estimation
Despite promise, PPG Hb estimation faces hurdles: motion artifacts distort signals during activity; skin tone variations cause bias (darker tones absorb more light); low perfusion in cold extremities; and overlapping absorption spectra.
| Challenge | Impact | Mitigation via Hb-PPG |
|---|---|---|
| Motion Artifacts | Signal noise | Multi-wavelength redundancy |
| Skin Pigmentation | Bias in low Hb | Diverse subject data for ML training |
| Low Perfusion | Poor SNR | BP references for correlation |
Enabling Machine Learning Innovations
Prior datasets are limited in scale or wavelengths; Hb-PPG fills this void for ML regression models (e.g., CNNs, LSTMs) predicting Hb from PPG features like AC/DC ratios, peaks.
Links to full paper and academic CV tips for researchers leveraging such data.
Applications in Wearables and UAE Healthcare
Imagine Apple Watch or Fitbit alerting to low Hb during exercise. Hb-PPG accelerates this for anemia screening in UAE clinics, maternity wards, or expatriate health checks. Integrates with Khalifa's Healthcare Engineering Innovation Group efforts.
Potential for telemedicine in remote emirates, aligning with UAE's Vision 2031 for advanced health tech.
Broader Implications for UAE Higher Education
Khalifa University's role highlights UAE's rise in global research, with Elgendi's interdisciplinary work bridging AI and medicine. Explore faculty positions at higher-ed-jobs/faculty or rate professors via Rate My Professor.
Collaborations like this dataset exemplify UAE-China ties in biotech.
Photo by mohamad azaam on Unsplash
Future Outlook and Calls to Action
Expect ML models trained on Hb-PPG to debut in prototypes soon. Researchers: Download the dataset, build models, cite the paper. For career advice in biomedical engineering, visit higher-ed-career-advice. Job seekers: Check university-jobs and higher-ed-jobs for UAE opportunities. Postdocs and faculty: higher-ed-jobs/postdoc.
This publication cements Khalifa University's legacy in transformative health tech.