Synthetic data is artificially generated information that looks and acts like the real thing. It lets researchers create large, controlled datasets to test ideas when getting real-world measurements would take years or cost a fortune.
Synthetic data is still a relatively young field, but it's quickly becoming a valuable tool in medical device development, helping teams test algorithms, model complex physiology, and move early-stage products forward even when live-patient data is scarce.
Hailey Trier, a postdoctoral machine learning engineer at AQMed, didn’t initially plan to work on synthetic data. It started as a tool of necessity.
During her PhD in computational neuroscience, Hailey was trying to model how people make decisions — how they weigh risk, uncertainty, and reward. But human behavior is so complex that developing a single model to account for every behavioral factor is nearly impossible. Furthermore, the more complicated a model is, the more data is required to test it. Brain imaging studies have small sample sizes and are limited by cost, time, and patient availability. To justify imaging the brain, Hailey needed to construct a valid model of human decision-making and generate a lot of data to support it.
That’s where synthetic data came in. "We needed to figure out if our model was even close to how humans really think, so we compared real human behavior with the behavior of our generative model,” Hailey said. By creating artificial populations of simulated subjects with a wide range of behavioral tendencies, she could test how well her models aligned with real-world behavior, providing the confidence and statistical power needed to apply the model to the human brain.
But generating data was only part of the challenge. She also had to ensure that these synthetic populations resembled not just a handful of real humans that inspired the model, but the diversity of a wider population. The work demanded constant iteration between simulation and validation, making sure that the model adequately captured the complexities of real data. After validating the model with synthetic data, Hailey and her team used brain imaging to link activity in a specific brain region to threat-related behaviors that have implications for mental health.
What began as a practical workaround became a deep expertise, one that Hailey would later carry into a very different field. Today, she’s applying those same skills to help solve a growing challenge in medical device development: how to build new medical device technologies when there’s almost no clinical data to work with. Data scarcity, as it’s called, complicates the development of medical technologies that could save lives.
At the center of her work is magnetocardiography (MCG), a cardiac diagnostic technology that captures the heart’s magnetic fields rather than its electrical signals. Unlike electrocardiograms (ECGs/EKGs), which rely on electrodes attached to the skin, MCG offers the potential for higher-resolution, contact-free measurements. The clinical promise of MCG is significant: potentially faster diagnosis of ischemia, non-invasive mapping of arrhythmias, and making fetal arrhythmia detection more widely available.
But for all its promise, MCG faces a core obstacle: there isn’t much data. Unlike ECG, which has been in clinical use for decades and has generated massive datasets, MCG is still emerging. There are too few clinical recordings to train algorithms or validate models. This creates a Catch-22 situation: the technology needs data to advance, but can’t easily generate that data without already being deployed at scale.
To break that cycle, Hailey turned again to synthetic data. Using physics-based models that calculate how electrical activity in the heart produces magnetic fields, the team transformed existing ECG recordings into simulated MCG signals. This synthetic data became a stand-in, allowing the team to train algorithms, test system performance, and begin building training sets even before large-scale real-world data becomes available. The goal is not to replace clinical data, but to fill a critical gap during early development, allowing device research to move forward while the collection of larger datasets through human subject research studies are underway.
The work draws directly on Hailey’s experience managing the risks of synthetic data generation. The models must account for physiological variability — differences in anatomy and disease states, for instance — to ensure the synthetic data remains relevant and accurate.
Although synthetic data comes with a lot of promise, there are plenty of ways to get it wrong. Machine learning models run on data, and synthetic data is a way to generate more of that “fuel.” It can really accelerate development, especially when real data is hard to get. But like any powerful fuel, you have to handle it carefully. If you cut corners or don’t validate it properly or treat it ethically, you can end up with models that fail in the real world. In healthcare, that means serious consequences for patients. That’s why the AQMed team thoroughly tests the validity of their simulations and how their models perform on real patient data.
Hailey’s path from academic neuroscience to medical device development reflects a broader shift happening across healthcare innovation. As new technologies push into areas where data is sparse or entirely missing, synthetic data is becoming a key tool for progress. For MCG, synthetic data may help unlock a diagnostic tool that could fundamentally change how heart disease is detected and managed.
Hailey’s journey — from modeling simulated minds to simulated hearts — offers a glimpse at how the careful use of synthetic data can help move promising ideas from theoretical possibility toward clinical reality. “Showing that we can use synthetic data to bring a product to market faster, and that [it results in] reliable and accurate diagnoses, will be a great proof point,” Hailey said. “I think a lot of other developers of medical technologies will be looking at [synthetic data].”
When thinking back on her journey from academia to industry, Hailey said: “I was doing more foundational science in my graduate study, which is really important…but I wanted to watch the science directly translate into a positive impact [in people’s lives],” she said. “I have found that here at SandboxAQ. And yeah, it feels good.”
In some ways, Hailey’s career has spanned both ends of the human experience — from modeling decisions in the brain to modeling rhythms in the heart. Different systems, different organs, but the same underlying challenge: how to use mathematics and equations to make sense of complex biology with limited real-world measurements. By bridging the brain and the heart, she’s quietly helping to open new frontiers for medical technology, one simulation at a time.
