Virtual twins bring models to life by coupling the interconnected systems of a single organ. Here, electrical pulses travel through a network of fibers to drive a heartbeat [right] and muscle tissue strains during the contractions [left].
One morning in May 2019, a cardiac surgeon stepped into the operating room at Boston Children’s Hospital more prepared than ever before to perform a high-risk procedure to rebuild a child’s heart. The surgeon was experienced, but he had an additional advantage: He had already performed the procedure on this child dozens of times—virtually. He knew exactly what to do before the first cut was made. Even more important, he knew which strategies would provide the best possible outcome for the child whose life was in his hands.
How was this possible? Over the prior weeks, the hospital’s surgical and cardio-engineering teams had come together to build a fully functioning model of the child’s heart and surrounding vascular system from MRI and CT scans. They began by carefully converting the medical imaging into a 3D model, then used physics to bring the 3D heart to life, creating a dynamic digital replica of the patient’s physiology. The mock-up reproduced this particular heart’s unique behavior, including details of blood flow, pressure differentials, and muscle-tissue stresses.
This type of model, known as a virtual twin, can do more than identify medical problems—it can provide detailed diagnostic insights. In Boston, the team used the model to predict how the child’s heart would respond to any cut or stitch, allowing the surgeon to test many strategies to find the best one for this patient’s exact anatomy.
That day, the stakes were high. With the patient’s unique condition—a heart defect in which large holes between the atria and ventricles were causing blood to flow between all four chambers—there was no manual or textbook to fully guide the doctors. The condition strains the lungs, so the doctors planned an open-heart surgery to reroute deoxygenated blood from the lower body directly to the lungs, bypassing the heart. Typically with this kind of surgery, decisions would be made on the fly, under demanding conditions, and with high uncertainty. But in this case, the plan had been tested in advance, and the entire team had rehearsed it before the first incision. The surgery was a complete success.
Such procedures have become routine at the Boston hospital. Since that first patient, nearly 2,000 procedures have been guided by virtual-twin modeling. This is the power of the technology behind the Living Heart Project, which I launched in 2014, five years before that first procedure. The project started as an exploratory initiative to see if modeling the human heart was possible. Now with more than 150 member organizations across 28 countries, the project includes dozens of multidisciplinary teams that regularly use multiscale virtual twins of the heart and other vital organs.
This technology is reshaping how we understand and treat the human body. To reach this transformative moment, we had to solve a fundamental challenge: building a digital heart accurate enough—and trustworthy enough—to guide real clinical decisions.
A father’s concern
Now entering its second decade, the Living Heart Project was born in part from a personal conviction. For many years, I had watched helplessly as my daughter Jesse faced endless diagnostic uncertainty due to a rare congenital heart condition in which the position of the ventricles is reversed, threatening her life as she grew. As an engineer, I understood that the heart was an array of pumping chambers, controlled by an electrical signal and its blood flow carefully regulated by valves. Yet I struggled to grasp the unique structure and behavior of my daughter’s heart well enough to contribute meaningfully to her care. Her specialists knew the bleak forecast children like her faced if left untreated, but because every heart with her condition is anatomically unique, they had little more than their best guesses to guide their decisions about what to do and when to do it. With each specialist, a new guess.
Then my engineering curiosity sparked a question that has guided my career ever since: Why can’t we simulate the human body the way we simulate a car or a plane?
At a visualization center in Boston, VR imagery helps the mother of a young girl with a complex heart defect understand the inner workings of her child’s heart. Dassault Systèmes
I had spent my career developing powerful computational tools to help engineers build digital models of complex mechanical systems, using models that ranged from the interactions of individual atoms to the components of entire vehicles. What most of these models had in common was the use of physics to predict behavior and optimize performance. But in medicine today, those same physics-based approaches rarely inform decision-making. In most clinical settings, treatment decisions still hinge on judgments drawn from static 2D images, statistical guidelines, and retrospective studies.
This was not always the case. Historically, physics was central to medicine. The word “physician” itself traces back to the Latin physica, which translates to “natural science.” Early doctors were, in a sense, applied physicists. They understood the heart as a pump, the lungs as bellows, and the body as a dynamic system. To be a physician meant you were a master of physics as it applied to the human body.
As medicine matured, biology and chemistry grew to dominate the field, and the knowledge of physics got left behind. But for patients like my daughter, that child in Boston, and millions like them, outcomes are governed by mechanics. No pill or ointment—no chemistry-based solution—would help, only physics. While I did not realize it at the time, virtual twins can reunite modern physicians with their roots, using engineering principles, simulation science, and artificial intelligence.
A decade of progress
The LHP concept was simple: Could we combine what hundreds of experts across many specialties knew about the human heart to build a digital twin accurate enough to be trusted, flexible enough to personalize, and predictive enough to guide clinical care?
We invited researchers, clinicians, device and drug companies, and government regulators to share their data, tools, and knowledge toward a common goal that would lift the entire field of medicine. The Living Heart Project launched with a dozen or so institutions on board. Within a year, we had created the first fully functional virtual twin of the human heart.
The Living Heart was not an anatomical rendering, tuned to simply replicate what we observed. It was a first-principles model, coupling the network of fibers in the heart’s electrical system, the biological battery that keeps us alive, with the heart’s mechanical response, the muscle contractions that we know as the heartbeat.
The Living Heart virtual twin simulates how the heart beats, offering different views to help scientists and doctors better predict how it will respond to disease or treatment. The center view shows the fine engineering mesh, the detailed framework that allows computers to model the heart’s motion. The image on the right uses colors to show the electrical wave that drives the heartbeat as it conducts through the muscle, and the image on the left shows how much strain is on the tissue as it stretches and squeezes. Dassault Systèmes
Academic researchers had long explored computational models of the heart, but those projects were typically limited by the technology they had access to. Our version was built on industrial-grade simulation software from Dassault Systèmes, a company best known for modeling tools used in aerospace and automotive engineering, where I was working to develop the engineering simulation division. This platform gave teams the tools to personalize an individual heart model using the patient’s MRI and CT data, blood-pressure readings, and echocardiogram measurements, directly linking scans to simulations.
Surgeons then began using the Living Heart to model procedures. Device makers used it to design and test implants. Pharmaceutical companies used it to evaluate drug effects such as toxicity. Hundreds of publications have emerged from the project, and because they all share the same foundation, the findings can be reproduced, reused, and built upon. With each application, the research community’s understanding of the heart snowballed.
Early on, we also addressed an essential requirement for these innovations to make it to patients: regulatory acceptance. Within the project’s first year, the U.S Food and Drug Administration agreed to join the project as an observer. Over the next several years, methods for using virtual-heart models as scientific evidence began to take shape within regulatory research programs. In 2019, we formalized a second five-year collaboration with the FDA’s Center for Devices and Radiological Health with a specific goal.
That goal was to use the heart model to create a virtual patient population and re-create a pivotal trial of a previously approved device for repairing the heart’s mitral valve. This helped our team learn how to create such a population, and let the FDA experiment with evaluating virtual evidence as a replacement for evidence from flesh-and-blood patients. In August 2024, we published the results, creating the first FDA-led guidelines for in silico clinical trials and establishing a new paradigm for streamlining and reducing risk in the entire clinical-trial process.
In 10 years, we went from a concept that many people doubted could be achieved to regulatory reality. But building the heart was only the beginning. Following the template set by the heart team, we’ve expanded the project to develop virtual twins of other organs, including the lungs, liver, brain, eyes, and gut. Each corresponds to a different medical domain, which has its own community, data types, and clinical use cases. Working independently, these teams are progressing toward a breakthrough in our understanding of the human body: a multiscale, modular twin platform where each organ twin could plug into a unified virtual human.
How a digital twin of the heart is constructed
A cardiac digital twin starts with medical imaging, typically MRI, CT, or both. The slices are reconstructed into the 3D geometry of the heart and connected vessels. The geometry of the whole organ must then be segmented into its constituent parts, so each substructure—atria, ventricles, valves, and so on—can be assigned their unique properties.
At this point, the object is converted to a functional, computational model that can represent how the various cardiac tissues deform under load—the mechanics. The complete digital twin model becomes “living” when we integrate the electrical fiber network that drives mechanical contractions in the muscle tissue.
Each part of the heart, such as the left ventricle [left], is superimposed with a detailed digital mesh to re-create its physiology. These pieces come together to form an anatomically accurate rendering of the whole organ [right].Dassault Systèmes
To simulate circulation, the twin adds computational models of hemodynamics, the physics of blood flow and pressure. The model is constrained by boundary conditions of blood flow, valve behavior, and vascular resistance set to closely match human physiology. This lets the model predict blood flow patterns, pressure differentials, and tissue stresses.
Finally, the model is personalized and calibrated using available patient data, such as how much the volume of the heart chambers changes during the cardiac cycle, pressure measurements, and the timing of electrical pulses. This means the twin reflects not only the patient’s anatomy but how their specific heart functions.
Building bigger cohorts with generative AI
When the FDA in silico clinical trial initiative launched in 2019, the project’s focus shifted from these handcrafted virtual twins of specific patients to cohorts large enough to stand in for entire trial populations. That scale is feasible today only because virtual twins have converged with generative AI. Modeling thousands of patients’ responses to a treatment or projecting years of disease progression is prohibitively slow with conventional digital-twin simulations. Generative AI removes that bottleneck.
AI boosts the capability of virtual twins in two complementary ways. First, machine learning algorithms are unrivaled at integrating the patchwork of imaging, sensor, and clinical records needed to build a high-fidelity twin. The algorithms rapidly search thousands of model permutations, benchmark each against patient data, and converge on the most accurate representation. Workflows that once required months of manual tuning can now be completed in days, making it realistic to spin up population-scale cohorts or to personalize a single twin on the fly in the clinic.
Second, enriching AI models’ training sets with data from validated virtual patients grounds the AI simulations in physics. By contrast, many conventional AI predictions for patient trajectories rely on statistical modeling trained on retrospective datasets. Such models can drift beyond physiological reality, but virtual twins anchor predictions in the laws of hemodynamics, electrophysiology, and tissue mechanics. This added rigor is indispensable for both research and clinical care—especially in areas where real-world data are scarce, whether because a disease is rare or because certain patient populations, such as children, are underrepresented in existing datasets.
Enabling in silico clinical trials
On the research side, the FDA-sponsored In Silico Clinical Trial Project that we completed in 2024 opened a new world for medical innovations. A conventional clinical trial may take a decade, and 90 percent of new drug treatments fail in the process. Virtual twins, combined with AI methods, allow researchers to design and test treatments quickly in a simulated human environment. With a small library of virtual twins, AI models can rapidly create expansive virtual patient cohorts to cover any subset of the general population. As clinical data becomes available, it can be added into the training set to increase reliability and enable better predictions.
The Living Heart Project has expanded beyond the heart, modeling organs throughout the body. The 3D brain reconstruction [top] shows major pathways in the brain’s white matter connecting color-coded regions of the brain. The lung virtual twin [middle] combines the organ’s geometry with a physics-based simulation of air flowing down the trachea and into the bronchi. And the cross section of a patient’s foot [bottom] shows points of strain in the soft tissue when bearing weight. Dassault Systèmes
Virtual twin cohorts can represent a realistic population by building individual “virtual patients” that vary by age, gender, race, weight, disease state, comorbidities, and lifestyle factors. These twins can be used as a rich training set for the AI model, which can expand the cohort from dozens to hundreds of thousands. Next the virtual cohort can be filtered to identify patients likely to respond to a treatment, increasing the chances of a successful trial for the target population.
The trial design can also include a sampling of patient types less likely to respond or with elevated risk factors, thus allowing regulators and clinicians to understand the risks to the broader population without jeopardizing overall trial success. This methodology enhances precision and efficiency in clinical research, providing population-level insights previously available only after many years of real-world evidence.
Of course, though today’s heart digital twins are powerful, they’re not perfect replicas. Their accuracy is bounded by three main factors: what we can measure (for example, image resolution or the uncertainty of how tissue behaves in real life), what we must assume about the physiology, and what we can validate against real outcomes. Many inputs, like scarring, microvascular function, or drug effects are difficult to capture clinically, so models often rely on population data or indirect estimation. That means predictions can be highly reliable for certain questions but remain less certain for others. Additionally, today’s digital twins lack validation for predicting long-term outcomes years in the future, because the technology has been in use for only a few years.
Over time, each of these limitations will steadily shrink. Richer, more standardized data will tighten personalization of the models. AI tools will help automate labor-intensive steps. And the collection of longitudinal data will improve the model’s ability to reliably predict how the body will evolve over time.
How virtual twins will change health care
Throughout modern medicine, new technologies have sharpened our ability to diagnose, providing ever-clearer images, lab data, and analytics that tell physicians what is presently happening inside a patient’s body. Virtual twins shift that paradigm, giving clinicians a predictive tool.
This “Living Lung” virtual-twin simulation shows strain patterns during breathing. Mona Eskandari/UC Riverside
Early demonstrations are already appearing in many areas of medicine, including cardiology, orthopedics, and oncology. Soon, doctors will also be able to collaborate across specialties, using a patient-specific virtual twin as the common ground for discussing potential interactions or side effects they couldn’t predict independently.
Although these applications will take some time to become the standard in clinical care, more changes are on the horizon. Real-time data from wearables, for example, could continuously update a patient’s personalized virtual twin. This approach could empower patients to understand and engage more deeply in their care, as they could see the direct effects of medical and lifestyle changes. In parallel, their doctors could get comprehensive data feeds, using virtual twins to monitor progress.
Imagine a digital companion that shows how your particular heart will react to different amounts of salt intake, stress, or sleep deprivation. Or a visual explanation of how your upcoming surgery will affect your circulation or breathing. Virtual twins could demystify the body for patients, fostering trust and encouraging proactive health decisions.
A new era of healing
With the Living Heart Project, we’re bringing physics back to physicians. Modern physicians won’t need to be physicists, any more than they need to be chemists to use pharmacology. However, to benefit from the new technology, they will need to adapt their approach to care.
This means no longer seeing the body as a collection of discrete organs and considering only symptoms, but instead viewing it as a dynamic system that can be understood, and in most cases, guided toward health. It means no longer guessing what might work but knowing—because the simulation has already shown the result. By better integrating engineering principles into medicine, we can redefine it as a field of precision, rooted in the unchanging laws of nature. The modern physician will be a true physicist of the body and an engineer of health.
Instagram Engineering
Shopify Engineering