Google Research AI flags high blood pressure from smartwatch pulse and motion data, no cuff calibration needed

A Google Research AI can detect signs of high blood pressure using a regular smartwatch, with no blood pressure cuff needed for calibration. In a 196-person study, the watch identified 65.8 percent of people with hypertension, compared to 55.3 percent for standard in-office measurement. The study was published in Nature, and Google described the work in a blog post.

How the system works

The AI relies on two sensors already built into modern watches: a pulse sensor and a motion sensor. A neural network learned from nearly 500,000 hours of sensor data. It compared short segments and worked out which ones belonged to the same person, picking up typical pulse and heartbeat patterns without any diagnoses to guide it.

The system then averages these patterns over several days and calculates a probability for high blood pressure.

Google's health model SensorFM also learns from smartwatch data.

Study results

The study covered 196 people across five US research sites. According to the reported results, the watch spotted hypertension about as well as a traditional cuff:

• Smartwatch AI: identified 65.8 percent of those with high blood pressure
• Standard in-office measurement: 55.3 percent

Product status

Google says the findings have been folded into the blood pressure trends feature in the Google Health app. The method itself is still a research project and does not replace a diagnosis. According to Google, the goal is to nudge people toward getting checked. Google also says more than half of people with high blood pressure do not know they have it.

The source does not give a model name for the hypertension detector, a parameter count, or a list of supported watch models.

What this means

The notable technical point is that the model needs no cuff calibration. It is trained by learning to match short sensor segments to the same person, with no diagnostic labels guiding that stage. Using only a pulse sensor and a motion sensor means the approach depends on hardware that regular smartwatches already include.

The reported comparison is also worth reading carefully. The watch identified 65.8 percent of people with hypertension, against 55.3 percent for standard in-office measurement, in a study of 196 people. That is a detection-rate comparison from a single study at five US sites, and Google itself frames the method as a research project that does not replace a diagnosis.

The practical value Google points to is screening: averaging patterns over several days and flagging a probability could prompt people who are unaware of their condition to get checked. The inclusion of the findings in the blood pressure trends feature in the Google Health app indicates the research has already reached a consumer product in some form.