Highlight
- AI-based Virtual Native Enhancement (VNE) MRI accurately detects myocardial scars without the need for gadolinium contrast agents.
- In a prospective multicenter blinded study, VNE achieved 94.4% diagnostic accuracy in myocardial infarction detection using high-quality images.
- Quantitative scar measurements by VNE strongly correlate with the current gold standard late gadolinium enhancement (LGE) imaging.
- VNE enabled clinical readers to forgo contrast administration in nearly 70% of patients without compromising diagnostic accuracy.
Study Background
Myocardial scarring following infarction is a critical determinant of patient prognosis, influencing arrhythmia risk, cardiac dysfunction, and subsequent management. Cardiovascular magnetic resonance (CMR) imaging with late gadolinium enhancement (LGE) remains the gold standard for visualizing myocardial scar due to its high spatial resolution and tissue characterization capabilities. However, gadolinium-based contrast agents entail several drawbacks including nephrotoxicity risk, prolonged imaging protocols, cost, and contraindications in patients with renal insufficiency or allergies.
The advent of artificial intelligence (AI) applications in medical imaging has fostered the development of contrast-free techniques for scar detection. Virtual native enhancement (VNE) leverages AI algorithms to synthesize scar-enhanced images from native cine and T1 mapping sequences, bypassing the need for contrast agents. As a transformative modality, VNE requires rigorous prospective validation to confirm reliability, reproducibility, and clinical utility across diverse populations and imaging centers.
Study Design
This prospective multicenter study involved two major cardiovascular imaging centers—Leeds (UK) and Fuwai (China)—and an independent AI development team at Oxford (UK). Consecutive patients undergoing CMR for myocardial scar assessment were enrolled to represent real-world clinical settings.
Scar presence was independently determined at Leeds and Fuwai using LGE imaging as the established reference standard. Oxford utilized only precontrast cine and T1 mapping images from the other sites to generate VNE images via a proprietary AI framework. Blinded to clinical information and LGE results, the Oxford team scored VNE images. Additionally, four experienced clinical readers performed independent blinded evaluation of matched VNE and LGE image slices to compare diagnostic concordance.
Primary endpoints included the diagnostic accuracy of VNE for myocardial infarction detection and quantitative correlation of scar size with LGE. Secondary outcomes assessed the potential of VNE to reduce the need for contrast administration without compromising clinical decision-making.
Key Findings
The study analyzed 136 matched CMR image datasets. Among confidently interpretable VNE images (n=107), myocardial infarction detection accuracy reached 94.4%, while considering all images resulted in an 87.5% accuracy. Quantitative scar burden by VNE correlated strongly with LGE measurements (Pearson’s R=0.90), with a mean difference in scar size of 3.2% (95% CI: -10.4% to +16.8%), demonstrating close agreement without systematic bias.
Further, the spatial agreement of infarcted myocardial territories between VNE and LGE was 90.0%, indicating robust anatomical fidelity. When clinical readers evaluated VNE images first, they deemed LGE unnecessary in approximately 69.7% of cases. In these cases, VNE achieved a mean diagnostic accuracy of 93.7%, comparable to LGE’s 93.9%, underscoring the potential for VNE to safely triage patients and reduce contrast use.
Safety data specifically related to contrast administration were not a primary endpoint but the contrast-free nature of VNE suggests inherent safety benefits by eliminating gadolinium exposure.
Expert Commentary
This study represents a landmark in the clinical translation of AI-powered, contrast-free myocardial scar imaging. The prospective blinded, multicenter design enhances the validity and generalizability of findings across different scanners, demographics, and clinical workflows. The high diagnostic accuracy and quantitative fidelity of VNE compared to LGE indicate AI’s potential to revolutionize conventional scar imaging paradigms.
Clinicians should note that although promising, VNE is currently complementary rather than a wholesale replacement for LGE. Certain low-quality or ambiguous VNE images might still necessitate conventional contrast enhancement for definitive diagnosis. Additionally, the impact of VNE on downstream clinical management, prognosis, and outcomes remains to be rigorously studied.
Future directions include further algorithm refinement for broader tissue pathologies, automated workflows embedded in routine CMR sequences, and cost-effectiveness analyses to support widespread adoption.
Conclusion
Artificial intelligence-based virtual native enhancement MRI offers a reliable, accurate, and non-contrast method for myocardial scar detection and quantification. By maintaining diagnostic concordance with LGE and enabling avoidance of gadolinium contrast in over two-thirds of chronic myocardial infarction assessments, VNE holds promise to transform clinical cardiac imaging.
Widespread implementation could improve patient safety, reduce costs, and streamline imaging protocols in routine cardiovascular care, pending larger scale validation and integration into clinical workflows.
Funding and Clinical Trial Registry
Funding sources and trial registration details were not specified in the available summary. Further consultation of the original publication (PMID: 42455101) is recommended for comprehensive information.
References
1. Zhang Q, Zhou D, Thompson P, et al. Myocardial Scar Assessment Using Artificial Intelligence-Powered Contrast-Free MRI: A Prospective Multicenter Study of Virtual Native Enhancement. J Am Coll Cardiol. 2026 Jul 7; PMID: 42455101.
2. Kim RJ et al. The Use of Contrast-Enhanced Magnetic Resonance Imaging to Identify Reperfused Myocardial Infarction. N Engl J Med. 2000;343(20):1445-1453.
3. Ferreira VM, Piechnik SK, et al. Non-contrast native T1 mapping for myocardial characterization: an update. JACC Cardiovasc Imaging. 2014;7(6): 665-681.
