Researchers at Queen Mary University of London have developed an artificial intelligence tool capable of creating synthetic but medically realistic models of heart scarring, potentially transforming how doctors plan treatment for atrial fibrillation (AF), the most common heart rhythm disorder. The study, published in Frontiers in Cardiovascular Medicine, shows how AI can overcome a fundamental barrier in personalised cardiac care, a chronic shortage of patient data, while supporting more accurate and efficient treatment decisions.
AF affects more than 1.4 million people in the UK and is often treated using catheter ablation, a procedure that deliberately scars the heart to interrupt erratic electrical signals. But with failure rates approaching 50 per cent, there is a pressing need for more tailored and predictable treatment approaches. The Queen Mary study points to a future where virtual replicas of diseased heart tissue, generated by AI, could support simulations of patient-specific interventions before any real-world procedure takes place.
Synthetic data and real consequences
At the heart of this innovation lies a sophisticated AI diffusion model trained on a limited set of 100 MRI scans from AF patients. Using these, the system generated an additional 100 synthetic patterns of fibrotic heart tissue, the stiff, scar-like patches that develop as a result of AF, ageing, or prolonged stress. These patterns were then integrated into 3D models of the heart to test the effectiveness of different ablation strategies across diverse anatomical and pathological scenarios.
“The AI-generated fibrosis distributions matched the real patient data with a high degree of fidelity,” said Dr Alexander Zolotarev, first author of the study. “That allowed us to simulate treatment responses in silico, with results that closely mirrored those derived from actual MRI data.”
The implications are twofold. First, it means researchers can now run extensive digital trials on heart models without breaching patient privacy or being constrained by limited imaging datasets. Second, it enables a deeper understanding of how structural variations in scar tissue might influence treatment success – something that remains difficult to predict using existing methods.
From data scarcity to simulation scale
While machine learning has shown promise in forecasting ablation outcomes, progress has been hindered by the limited availability of high-resolution, labelled clinical data. LGE-MRI scans are a critical resource in this space, but their acquisition is time-consuming, expensive, and often restricted due to patient confidentiality.
By generating synthetic fibrosis that replicates the complexity of real tissue, the new AI model sidesteps these bottlenecks. It opens the door to large-scale simulation studies that could help refine treatment protocols and, eventually, assist clinicians in personalising ablation strategies for individual patients.
“This isn’t about replacing clinical judgement,” said Zolotarev. “It’s about enhancing it, giving doctors the ability to rehearse and optimise their approach for each patient using a digital heart twin that accurately reflects their unique condition.”
Dr Caroline Roney, lead author of the study and recipient of a UKRI Future Leaders Fellowship, describes the work as a foundational step toward truly personalised medicine in cardiology. “What we’ve demonstrated is a way to safely and effectively extend the reach of cardiac modelling,” she said. “Our ultimate aim is to support large-scale in silico trials and bring patient-specific simulation into everyday clinical decision-making.”
A digital twin for every patient
The vision of a digital twin, a precise virtual counterpart of a patient’s heart, has long been seen as a promising avenue for improving treatment outcomes. However, the challenge of populating these models with accurate, granular data has slowed their development.
By using AI to produce clinically useful synthetic fibrosis maps, the Queen Mary team not only addresses a technical barrier, but also offers a route to protect patient anonymity while retaining the physiological fidelity essential for reliable modelling.
Given the scale of AF and the high failure rate of existing treatment options, the technology arrives at a critical juncture. It suggests a future where cardiologists can stress-test therapies against digital replicas before entering the catheter lab, minimising guesswork and maximising efficacy.
In an era where AI’s role in healthcare is often viewed through the lens of automation or decision-making, this research offers a subtler, yet potentially more transformative vision, AI as a silent partner in the background, enriching human expertise by filling in the data gaps that medicine has long struggled to close.
