A digital replica of a child's failing heart, built from MRI and CT scans, allowed a cardiac surgeon at Boston Children's Hospital to rehearse a high-risk open-heart procedure dozens of times virtually before performing it in May 2019 — and the surgery was a complete success.

That operation was the first clinical use of a technology called a virtual twin developed through the Living Heart Project, a research initiative launched in 2014 and built on industrial-grade simulation software from Dassault Systèmes. Since that first case, the hospital has used virtual-twin modelling to guide nearly 2,000 procedures, according to the project's founder. The project now includes more than 150 member organisations across 28 countries.

How a Digital Heart Actually Works

A cardiac virtual twin begins with standard medical imaging — typically MRI and CT scans — which are reconstructed into a three-dimensional geometry of the heart and its connected vessels. Engineers then segment that geometry into constituent parts: atria, ventricles, valves, and surrounding tissue. Each substructure receives its own physical properties.

The model becomes functional when two systems are integrated: the electrical fibre network that triggers muscle contractions, and the mechanical response of the tissue itself. A third layer adds haemodynamics — the physics of blood flow, pressure, and vascular resistance. Finally, the twin is personalised using patient-specific data, including chamber volume changes across the cardiac cycle, pressure readings, and the timing of electrical pulses.

The result is not a static anatomical illustration. It is a predictive simulation of how one specific patient's heart will behave under specific conditions — including surgical interventions that have not yet happened.

Virtual twins provide clinicians with a predictive tool to anticipate how a patient's heart will respond to specific conditions and interventions.

A Personal Mission Behind the Science

The project's origins are partly personal. The founder, an engineer specialising in computational simulation, had for years watched his daughter Jesse face diagnostic uncertainty due to a rare congenital condition in which the positions of the ventricles are reversed. Her specialists understood the prognosis but, because every heart with her condition is anatomically unique, relied largely on clinical judgment rather than patient-specific prediction.

That experience prompted a question that has since shaped the project: why can't the human body be simulated the way engineers simulate aircraft or automobiles? Physics-based modelling is standard practice in aerospace and automotive design, where digital prototypes are stress-tested thousands of times before any physical component is built. In most clinical settings, by contrast, treatment decisions still rely on static 2D images, statistical guidelines, and retrospective studies.

The Living Heart Project set out to close that gap, inviting researchers, clinicians, device manufacturers, pharmaceutical companies, and government regulators to share data and knowledge toward a common model. Within its first year, the collaboration produced the first fully functional virtual twin of the human heart.

From Single Patients to Population-Scale Trials

The clinical application is only part of the ambition. In 2019, the U.S. Food and Drug Administration formally joined the project as a collaborator, with a specific goal: to use virtual-heart models to recreate a pivotal clinical trial of a previously approved device for repairing the heart's mitral valve, testing whether virtual evidence could substitute for evidence gathered from real patients.

In August 2024, the project published those results, and the FDA released what it described as the first regulatory guidelines for in silico clinical trials — simulated trials run entirely on virtual patient populations. A conventional clinical trial can take a decade to complete, and 90 percent of new drug treatments fail during that process, according to the project's published findings. Virtual cohorts built from individual twins, scaled using generative AI, could allow researchers to test interventions on populations of hundreds of thousands of simulated patients before a single human subject is enrolled.

Generative AI plays two distinct roles in that scaling process. First, machine learning algorithms integrate the patchwork of imaging data, sensor readings, and clinical records required to build a high-fidelity twin — a workflow that previously took months of manual tuning and can now be completed in days, according to the project team. Second, AI models trained on validated virtual patients are grounded in the physics of haemodynamics and tissue mechanics, reducing the risk of predictions that drift beyond physiological reality — a known weakness of purely statistical models trained on retrospective datasets.

Limitations the Project Openly Acknowledges

The technology has clear constraints. Model accuracy is bounded by what can be measured — image resolution, the uncertainty of real tissue behaviour — by the assumptions required to fill gaps in patient data, and by the availability of real-world outcomes against which predictions can be validated. Factors such as scarring, microvascular function, and drug interactions are difficult to capture clinically, so models often rely on population-level estimates.

Importantly, today's digital twins lack validation for predicting long-term outcomes years into the future, because the technology has only been in widespread clinical use for a few years. The project's published work acknowledges that predictions can be highly reliable for certain questions and considerably less certain for others.

The development team expects those limitations to shrink as richer, more standardised patient data tightens personalisation, AI tools automate labour-intensive modelling steps, and longitudinal data accumulates to test long-range predictions.

Beyond the Heart

Following the cardiac template, the Living Heart Project has expanded to develop virtual twins of the lungs, liver, brain, eyes, and gut. Each organ twin corresponds to a different medical community, with its own data types and clinical use cases. The long-term goal, according to the project, is a modular virtual human — a unified platform in which each organ twin connects to the others, allowing specialists across disciplines to model interactions between systems rather than treating organs in isolation.

Early applications are already appearing in oncology, where clinicians are modelling tumour growth and the body's response to different therapies, and in orthopaedics, where surgeons are personalising implants by accounting for whole-body kinematics rather than local anatomy alone.

Looking ahead, the project envisions virtual twins continuously updated by data from wearables — giving patients a tool to visualise how their own physiology responds to salt intake, stress, sleep, or a proposed surgical intervention, and giving clinicians a real-time data feed to monitor progress between appointments.

What This Means

For patients facing complex or rare conditions — particularly children, who are chronically underrepresented in clinical trial datasets — virtual twins offer the prospect of treatment decisions grounded in their specific physiology rather than population averages, offering a new approach to personalised medicine in practice.