PINN captures arterial flow from wearable watch

A research team led by Benjamin Sanchez said a cuffless smartwatch using electrical bioimpedance and a physics-informed neural network can estimate arterial blood velocity and blood pressure from the wrist. The work was posted as an arXiv preprint on Dec. 31, 2025, and a version titled “Cuffless hemodynamic monitoring with physics-informed machine learning models” was published by Nature Communications on May 14, 2026. (pmc.ncbi.nlm.nih.gov) A social-media demo shared on May 18 highlighted one piece of the validation: peak systolic flow velocity from the watch compared with Doppler ultrasound. The claim is narrower than “a watch measures everything.” The paper says the device uses real-time electrical bioimpedance, or BioZ, and a signal-tagged physics-informed neural network that incorporates fluid-dynamics constraints to estimate blood pressure and both radial and axial blood velocity. (arxiv.org) The authors said they tested the approach in healthy people and in patients with hypertension and cardiovascular disease across outpatient and intensive-care settings. ### What exactly did the watch measure? The authors said the watch records electrical bioimpedance at the wrist, a signal that changes as blood volume and flow change through the cardiac cycle. (arxiv.org) Instead of treating that waveform as a generic proxy, the model links the signal to hemodynamics through a physics-based framework, according to the abstract and paper summary. Peak systolic flow velocity is the specific quantity shown in the May 18 demo. In practical terms, that is the highest blood-flow velocity during the heart’s pumping phase, and Doppler ultrasound is a standard noninvasive way to measure blood-flow velocity for comparison. (pmc.ncbi.nlm.nih.gov) The team’s post said the watch-derived and Doppler-derived values showed quantitative agreement. ### Why use a physics-informed neural network here? Nature Communications’ summary said existing cuffless blood-pressure wearables often rely on pulse-wave analysis or pulse arrival time, approaches the authors argued can be undermined by physiological and experimental confounders. (pmc.ncbi.nlm.nih.gov) The Sanchez team said its model instead embeds fluid-dynamics principles directly into the learning process. That matters because the system is not only fitting patterns from labeled data. The abstract says the network is constrained by a mechanistic link between the BioZ signal and cardiovascular dynamics, which the authors said enables calibration-free estimation of blood pressure and blood velocity. (pmc.ncbi.nlm.nih.gov) ### How was the watch checked against ultrasound? The May 18 demonstration focused on Doppler ultrasound as the reference measurement for peak systolic flow velocity. The social post described a direct comparison between watch-derived velocity and Doppler-derived velocity in study participants, rather than a looser correlation with a downstream metric. (nature.com) The broader paper reports testing in multiple groups and settings, including healthy individuals at rest, after physical activity, during physical and autonomic challenges, and in patients with hypertension and cardiovascular disease. The abstract does not provide all performance tables in the search snippets, but it does state that the approach was “successfully tested” across those settings. (pmc.ncbi.nlm.nih.gov) ### Is this already a consumer-ready blood-pressure watch? The paper describes a research device and validation study, not a commercial product launch. The Nature Communications page and arXiv record present it as a scientific report on cuffless hemodynamic monitoring, and the preprint version runs 225 pages with supplementary figures, tables and videos. (pmc.ncbi.nlm.nih.gov) A 2026 scientific statement in the European Journal of Preventive Cardiology said cuffless blood-pressure devices are promising but still face validation and implementation challenges. That places the Sanchez team’s results inside a broader field that is still working through accuracy, standards and clinical use. (pmc.ncbi.nlm.nih.gov) ### Where can readers see the next layer of evidence? The Nature Communications article published May 14 and the arXiv preprint 2601.00081 contain the methods, author list and supplementary material for the smartwatch system. The authors listed on the preprint include Henry Crandall, Tyler Schuessler, Ramakrishna Mukkamala, Alfred K. (nature.com) Cheung, Stavros G. Drakos, Christel Hohenegger, Braxton Osting and Benjamin Sanchez. The next step for readers is to review the full paper’s validation tables and supplementary videos, where the team said it documented blood-pressure and blood-velocity estimation across healthy volunteers and cardiovascular patients. (academic.oup.com) (arxiv.org) (nature.com)