Passive heart-rate monitoring during smartphone use in everyday life

Article Passive heart-rate monitoring during smartphone use in everyday life https://doi.org/10.1038/s41586-026-10507-6 Received: 14 March 2025 Accepted: 8 April 2026 Shun Liao1,5, Paolo Di Achille2,5, Jiang Wu1,5, Silviu Borac1,5, Jonathan Wang1,5, Xin Liu3,5, Eric S. Teasley1,5, Lawrence Cai1, Yuzhe Yang1, Yun Liu1, Daniel McDuff3, Hao-Wei Su2, Brent Winslow1, Anupam Pathak1, Mark Malhotra1, Shwetak Patel3,4, James A. Taylor3, Jameson K. Rogers1,6 & Ming-Zher Poh2,6 ✉ Published online: xx xx xxxx Open access Check for updates Resting heart rate (RHR) is a key biomarker of cardiovascular health and mortality1–3, but passively tracking it longitudinally generally requires a wearable device, limiting its availability. Here we present passive heart-rate monitoring (PHRM), a deep-learning system that uses facial video-based photoplethysmography for passive measurements of heart rate (HR) and RHR during everyday smartphone interactions. Our system was developed using 192,353 videos from 485 participants and validated on 162,546 videos from 211 participants in laboratory and free-living conditions, representing, to our knowledge, the largest validation study of its kind. PHRM outperformed state-of-theart methods on our benchmarks. Compared with reference electrocardiograms, PHRM achieved a mean absolute percentage error (MAPE) lower than 10% for HR measurements across three skin-tone groups of light, medium and dark pigmentation, meeting industry accuracy standards; MAPE for each skin-tone group was non-inferior versus the others. Daily RHR measured by PHRM had a mean absolute error of less than five beats per minute, compared with a wearable HR tracker, and was associated with known risk factors for cardiovascular disease. These results highlight the potential of smartphones for enabling passive and equitable monitoring of heart health. To facilitate further research, we publicly release a large, annotated smartphone video dataset along with a pre-trained HR model. Heart rate (HR) is an important and dynamic vital sign that is influenced by numerous inputs4, and resting heart rate (RHR) is recognized as a biomarker and prognostic factor for overall mortality1–3. Longitudinal increases in RHR are associated with higher mortality and adverse cardiovascular events5–7. Measurement of RHR conventionally requires a sustained period of rest, which limits the practicality of evaluating long-term trajectories. However, the sensitivity of HR to various factors suggests that the cardiovascular system is better assessed through multiple daily measurements than through brief, standardized clinic-based measurements8–10. Daily average HR has been shown to be a strong independent predictor of all-cause mortality11, even more so than clinic-measured RHR, and consumer wearable devices typically derive a daily RHR by passively aggregating HR measurements during periods of rest throughout the day12. Daily RHR monitoring can provide insights into cardiovascular health and detect physiological changes linked to fitness levels or illness13–15. Nonetheless, the adoption of consumer wearables, while growing, remains limited, especially among those who are most likely to benefit from these health-monitoring technologies16. Given that smartphones are already ubiquitous—owned by 90% of US adults and 69% of people globally17, and used 144 times daily, on average18—they offer an attractive alternative for opportunistic HR measurements across the day during normal phone use. The blood volume pulse can be measured from a distance using a technique called video-based remote photoplethysmography (rPPG)19,20, which can measure HR21–24 and screen for irregular rhythms, such as atrial fibrillation25, through smartphone cameras. However, existing rPPG studies have small sample sizes, are limited to controlled environments and face generalizability issues in real-world conditions. Crucially, the accuracy of current rPPG methods is known to drop significantly for darker skin tones, owing to an increased concentration of melanin26. Similar concerns apply to other PPG-based devices, such as pulse oximeters, which has led to scrutiny and calls for diversity in validation studies from health governing bodies like the US Food and Drug Administration (FDA) and the UK National Health Service (NHS)27,28. Furthermore, as previous rPPG studies have mainly involved active HR measurements in situated conditions, there remains a need to address passive HR measurements during everyday phone use under unconstrained, free-living conditions. In this study, we present a smartphone-based deep-learning system that enables passive measurements of both HR and daily RHR in the background during normal phone use (collectively referred to as passive heart-rate monitoring; PHRM). Compared with previous work, our system provides several advances. First, we validate its performance in a prospective study on a large and diverse set of videos (more than Google Research, Mountain View, CA, USA. 2Google Research, Cambridge, MA, USA. 3Google Research, Seattle, WA, USA. 4University of Washington, Seattle, WA, USA. 5These authors contributed equally: Shun Liao, Paolo Di Achille, Jiang Wu, Silviu Borac, Jonathan Wang, Xin Liu, Eric S. Teasley. 6These authors jointly supervised this work: Jameson K. Rogers, Ming-Zher Poh. ✉e-mail: 1 Nature | www.nature.com | 1 Article a Screen unlock Input video Video stabilization Face crop and resize Frame difference Neural network Confidence gating HRt1 HRt2 HRt3 … HRtn HR measurements HRt1 HRt3 … HRtn Aggregation and Kalman filter Daily RHR b Train n = 102,982 (82 participants) Tune n = 69,948 (46 participants) In-laboratory study 5 n = 1,731 videos (104 participants) Datasets Free-living study n = 326,745 videos (235 participants) Test n = 160,815 (107 participants) Optimize Train DNN and RHR algorithm Evaluate Final model PHRM Fig. 1 | Overview, development and validation of the PHRM system. a, In our research study with consented participants, after a screen-unlock event, PHRM passively captures, processes and analyses 8-s facial video clips using a deep neural network (DNN) to estimate HR and associated prediction confidence to determine whether the measurement is valid. To compute daily RHR, PHRM 162,000), collected in laboratory conditions as well as in free-living, real-world conditions using participants’ personal phones. Second, our system meets industry accuracy standards and achieves prespecified non-inferiority targets for people of all skin tones, demonstrating its potential for equitable HR monitoring. PHRM outperformed state-of-the-art methods on our benchmarks. Third, we show that PHRM-derived daily RHR also achieves prespecified levels of accuracy and is associated with well-established cardiovascular health metrics and risk factors. Finally, we publicly release both a pre-trained HR model and a large and diverse smartphone video dataset comprising all skin pigment (...truncated)