TOPLINE

An AI method applied virtual native enhancement (VNE) to precontrast cardiac magnetic resonance images and accurately identified myocardial infarction (MI) in most cases without using contrast agents, with close agreement with the traditional contrast-based method.

METHODOLOGY

• Researchers conducted a prospective validation to assess whether an AI method could detect and quantify myocardial scar without the gadolinium contrast routinely used for late gadolinium enhancement (LGE).
• They collected cardiovascular magnetic resonance data of 136 patients from two sites in the UK and China between 2023 and 2024. The mean age was in the early sixties, and they were predominantly men.
• Researchers used precontrast cine and T1-mapping scans and applied a pretrained generative AI model to generate VNE images without using gadolinium.
• Blinded expert readers compared VNE images against standard LGE scans, which were independently scored by separate site teams.
• Researchers reported diagnostic accuracy for the detection of infarcts, scar burden, percentage of the heart wall’s thickness replaced by scar tissue (transmurality), and agreement and correlation between VNE and LGE.

TAKEAWAY

• On 107 confident images, VNE achieved a per-patient accuracy of 94.4% for identifying MI (sensitivity, 92.5%; specificity, 97.5%). The accuracy was 87.5% across all 136 cases (sensitivity, 94.1%; specificity, 76.5%).
• The scar burden estimated by VNE correlated strongly with LGE (correlation coefficient [R], 0.90), with a mean difference of 3.2%.
• Scar transmurality quantified by VNE correlated strongly with LGE (R, 0.82), with a mean difference of 6.3%.
• On the basis of reader judgment, VNE could have allowed LGE to be omitted in about 69.7% of cases. In those cases, VNE achieved an average diagnostic accuracy of 93.7%, which closely matched that of LGE (93.9%).

IN PRACTICE

“High-quality VNE may triage the need for LGE and obviate its use in more than two thirds of referrals for chronic MI scar assessment without compromising diagnostic accuracy. These findings support that VNE that incorporates quality control holds promise to transform cardiac imaging and noninvasive myocardial scar assessment in the near future,” the researchers of the study wrote.

SOURCE

The study was led by Qiang Zhang, PhD, University of Oxford, Oxford, England. It was published online on July 15 in JACC.

LIMITATIONS

The VNE model was trained and validated using LGE images, which may have introduced bias into the reported accuracy. The image analyst was allowed to use both VNE and LGE images when placing reference regions; thus, scar quantification was not fully independent. Most infarcts were anterior, which limited the generalizability of the findings to inferior and lateral infarcts.

DISCLOSURES

Several authors reported receiving research grants, fellowships, or doctoral support from charitable medical research foundations, government or national research agencies, universities, and hospital or clinical research center funding. One author reported receiving support through a studentship partially funded by an industry, some reported having research agreements with imaging equipment vendors, and some reported holding university-managed patents. Detailed disclosures are available in the original article.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.