ORFAN-MAESTRIA: AI Predicts Cardioembolic Stroke, Atrial Fibrillation from CT Scans

The ORFAN-MAESTRIA trial has demonstrated that AI can detect and predict cardioembolic stroke and atrial fibrillation (AF) from routine cardiac computed tomography (CT) scans of patients.1

These data were presented at the European Society of Cardiology (ESC) Congress 2026 in Munich, Germany, by Charalambos Antoniades, MD, PhD, director of the Acute Multidisciplinary Imaging & Interventional Centre and BHF chair of cardiovascular medicine at the University of Oxford and founder and chief scientific officer at Caristo Diagnostics.1

“The main key message of the study is that we have atrial myopathy that can be detected on routine CT angiography,” Antoniades told HCPLive in an exclusive interview. “These atrial myopathy signals predict stroke independently from atrial fibrillation.”

ORFAN-MAESTRIA Structure

Coronary computed tomography angiography (CCTA) is the first investigative step for patients with chest pain. However, its widespread application has revealed a population without CAD with unclear prognosis and management. To this end, the ORFAN and MAESTRIA programs were enacted to determine whether the use of the perivascular fat attenuation index (FAI) Score could enable risk prediction and guide management in this population.2

ORFAN-MAESTRIA was a multicenter cohort study of 81,127 adult patients undergoing coronary CT angiography (CCTA) across 10 European centers. Antoniades and colleagues trained a fully automated pipeline for left atrium (LA) and peri-atrial adipose tissue (PAAL) segmentation via a deep learning model. The team extracted volumetric, morphologic, and higher-order radiomic features from the LA, PAAT, epicardial adipose tissue (EAT), and the whole heart – AI-based ensemble models generated the Atriomic Stroke Algorithm (ASA) and the AF prediction algorithm (AFA), trained in 40,513 patients from 7 UK centers and externally validated in 40,614 patients from an additional 3.1

Predicting Stroke With AI

Over 10 years of follow-up, Antoniades and colleagues found that 648 patients developed cardioembolic stroke, and 3463 developed AF. ASA successfully stratified stroke risk, identifying intermediate- and high-risk groups with 4.4-fold and 17.1-fold higher risks of cardioembolic stroke, respectively, versus low-risk (P <.001). Of the top 10 radiomic features included in the model, 7 were derived from the peri-atrial space, and only 3 from the LA itself.1

AFA also identified individuals at significantly elevated AF risk, with 3.1-fold and 11.5-fold higher AF incidence in intermediate- and high-risk groups versus low risk (P <.001). All 10 radiomic features contributing to the AFA model were derived from the LA. Additionally, AFA use improved prediction over the age- and sex-based model.1

Ultimately, Antoniades and colleagues determined that fully automated phenotyping of the peri-LA space enables detection of LA myopathy and predicts cardioembolic stroke from routine CCTA. However, AF prediction is achieved via radiomic phenotyping of the LA itself, dissociating the 2 conditions early on in their development.1

“At the moment, we predict stroke very accurately. The area under the curve, which is one of the metrics that they use to evaluate the quality of these models, is very high for this type of prediction,” Antoniades said. “Now, the predictions for stroke seem to be driven by inflammation in the atrium, at least based on the findings that we have. The treatments for stroke may not be only anticoagulants or antiplatelets in the future; it could be a combination, but this remains to be seen.”

Editors’ Note: Antoniades reports disclosures with AstraZeneca, Caristo Diagnostics, Lexicon, Novo Nordisk, Sanofi, Silence Therapeutics, and others.

References

1. Antoniades C. Using AI to detect atrial myopathy and predict cardioembolic stroke from routine cardiac CT: the ORFAN-MAESTRIA study. Presented at the European Society of Cardiology (ESC) Congress 2026, Munich, Germany. August 28-31, 2026.

2. Chan K, Wahome E, Tsiachristas A, et al. Inflammatory risk and cardiovascular events in patients without obstructive coronary artery disease: The ORFAN multicentre, Longitudinal Cohort Study. The Lancet. 2024;403(10444):2606-2618. doi:10.1016/s0140-6736(24)00596-8