Apple study shows potential of modern heart sensor technology Apfelpatient

Apple has been exploring for years how optical sensors in the Apple Watch can provide more insights into heart health. A new study now shows how artificial intelligence could extract deeper heart data from simple PPG signals. These findings build upon features Apple has already implemented while also offering a glimpse into potential future developments.

With watchOS 26, Apple introduced high blood pressure notifications on the Apple Watch. The feature uses the optical heart sensor to monitor how blood vessels respond to heartbeats over a 30-day period. The algorithm looks for consistent signs of high blood pressure and sends notifications when patterns are detected. Apple points out that this is not a medical diagnostic tool and that it won't identify all cases. Nevertheless, the company expects to uncover over one million previously undetected cases in the first year.

The key is to avoid evaluating isolated measurements and instead analyze long-term trends in the background. This is precisely the focus of Apple's new study, which demonstrates how AI could extract more information from optical data.

More data from the optical sensor

The study doesn't explicitly mention the Apple Watch. As with other publications on Apple's Machine Learning Research Blog, the work focuses on fundamental research and potential technological approaches. Titled "Hybrid Modeling of Photoplethysmography for Non-Invasive Monitoring of Cardiovascular Parameters," the researchers propose a hybrid approach that combines hemodynamic simulations and unlabeled clinical data to estimate cardiovascular biomarkers directly from PPG signals.

PPG stands for photoplethysmography and is the same optical sensor technology used in the Apple Watch. The signals differ slightly from those of a finger sensor, but it's all based on the same method: optically detecting changes in blood volume.

How Apple built the model

The research team created two datasets. One contains simulated arterial pressure waveforms, known as APWs. The second comprises real measurements in which PPG and APW data were simultaneously acquired. Subsequently, a generative model was trained to learn how to correlate PPG signals with the corresponding pressure waves.

The model generates several plausible APW waveforms from a PPG segment. These are further processed in a second model. This second model was trained with simulated APW data to which known cardiovascular parameters were assigned. These include stroke volume and cardiac output. Based on this, the model can derive the corresponding cardiac values for each PPG segment. The results are then averaged to obtain a final estimate, including an uncertainty value.

The results in detail

A new dataset was used for the analysis. It contains PPG and APW signals from 128 patients who underwent non-cardiac surgery and had available cardiovascular biomarkers. As the data were processed, the method proved able to accurately track temporal trends in stroke volume and cardiac output. While the absolute values were not exact, the trend representation was reliable.

Compared to traditional techniques, the hybrid AI method delivered better results. The study shows that combining simulation and real-world measurements can significantly increase the predictive power of a simple optical sensor. At the same time, this approach avoids the need for invasive, expensive, and extensive annotations.

What the researchers see next

In their conclusion, the team emphasizes that the hybrid approach is promising because it integrates physical knowledge into the model and does not rely solely on labeled data. This makes the method more robust, as labeled medical data is often difficult to obtain. Nevertheless, the researchers still see challenges. Predicting absolute values, in particular, remains difficult. They see this as an important task for future work.

Further improvements could arise from alternative generative models or new architectural variations of the networks. Furthermore, the same learning strategy used here with finger PPG could be transferred to wearable PPG sensors. In the long term, this would enable passive and continuous monitoring of cardiac biomarkers.

Potential of optical sensors for future Apple applications

Even though it remains unclear whether Apple will ever integrate these technologies into products like the Apple Watch, the study clearly demonstrates the significant potential of existing sensor technology. Optical sensors could provide far more than just heart rate or oxygen saturation in the future. With AI-powered models, it would be possible to monitor complex cardiac parameters in everyday life. This would further strengthen Apple's role in the healthcare sector and could make a significant contribution to the early detection and monitoring of heart problems. (Image: Shutterstock / Gabo_Arts)