By Dave DeFusco
What if you could see a forecast of your future health鈥攏ot just as numbers on a chart, but as a lifelike digital version of yourself that changes over time based on what you eat and how you live? That idea is at the heart of a research paper, 鈥淭oward Personalized Health Forecasting: An Intelligent Digital Twin Model for Diet-Influenced Biomarker Trajectories,鈥 presented in January by researchers in the Katz School鈥檚 Department of Graduate Computer Science and Engineering at the 40th Annual AAAI Conference on Artificial Intelligence in Singapore, one of the top AI conferences in the world.
The study introduces a prototype 鈥渄igital twin鈥 system designed to show how a person鈥檚 diet can shape their health years into the future. A digital twin is a virtual copy of a real person or object. In engineering, digital twins are often used to model airplanes or buildings. In healthcare, researchers are exploring how digital twins could help people better understand their own bodies. In this case, the researchers created digital twins of people that reflect changes in health linked to diet, weight, blood pressure, cholesterol and other medical measures called biomarkers.
鈥淥ur goal was to make long-term health data understandable and personal,鈥 said Ashikur Nobel, lead author of the paper and a Ph.D. student in computer science in the Department of Graduate Computer Science and Engineering. 鈥淚nstead of showing people complicated graphs or statistics, we wanted to show what those numbers actually mean for their body over time.鈥
The system works by combining three main parts. First, it uses a computer model that looks for patterns in long-term health data. The model studies how diet quality, along with information like age, weight and medical history, is connected to changes in biomarkers over several years. Unlike simpler models that look at one factor at a time, this one can capture how many factors interact in complex ways.
Second, the system translates those predictions into plain language. Rather than saying something like 鈥渨aist circumference increases by three units,鈥 the system produces descriptions that are easier to grasp, such as changes in body shape or health risk trends. These descriptions are meant to sound more like how a person would naturally talk about their health.
Finally, the system turns those descriptions into 3D digital avatars. These digital twins visually change as the predicted health conditions change. For example, shifts in weight, body shape or overall health status can be seen directly on the virtual body. The result is a moving, visual story of a person鈥檚 health journey.
This research builds on a broader effort at the Katz School to use artificial intelligence to understand diet and health over time. The research is funded by the National Institutes of Health through the 鈥溾 grant and additional NIH/NIDDK awards totaling over $3.1 million, led by Julia Fang, director of the Katz School鈥檚 M.S. in Artificial Intelligence, to develop an artificial intelligence platform that can identify meaningful patterns in long-term dietary data. The system uses machine learning to compare eating habits at both the individual and population levels, helping researchers generate new evidence that can inform future dietary guidelines. These efforts highlight how AI can transform large, complex nutrition datasets into practical health insights.
To test the digital twin idea, the researchers used long-term data from more than 400 people, collected at several points over up to six years. The data included diet quality scores based on what people ate, as well as medical measures like blood pressure, cholesterol levels and waist size. The model learned from most of the data and then tested its predictions on the rest to determine its accuracy.
The results were promising. In many cases, the system predicted future health measurements more accurately than several widely used computer models. While the researchers stress that this is still a prototype, the findings suggest that digital twins could one day become a powerful tool for personalized health forecasting.
鈥淭his work shows how artificial intelligence can move beyond black-box predictions,鈥 said Fang. 鈥淏y translating complex data into language and visuals that people can understand, this approach has the potential to make health guidance more personal, more intuitive and more useful.鈥
Although the digital twins in this study were created using past data, the system is designed with the future in mind. The same structure could support real-time updates, allowing people to explore 鈥渨hat if鈥 scenarios, such as how changing their diet today might affect their health years down the road.
The researchers also see potential benefits for doctors and healthcare professionals. Visual digital twins could make it easier to explain health risks, track progress and tailor advice to each individual鈥檚 lifestyle and needs.
There are still limitations to address, including expanding the system to more diverse populations and improving how the 3D models are evaluated. The study, however, offers a glimpse of a future where health forecasting is not just about numbers, but about understanding yourself.
鈥淧eople make better decisions when they can see the consequences clearly,鈥 said Jacob Matos, co-author of the study and a master鈥檚 student at UMass Dartmouth. 鈥淚f we can help someone visualize how their daily choices shape their long-term health, that鈥檚 a powerful step toward prevention and personalized care.鈥
