🧭 REBEL Rundown
📌 Key Points
- ♥️ Queen of Hearts AI outperformed both emergency physicians and cardiologists when interpreting challenging STEMI-equivalent and STEMI-mimic ECGs.
- ⚡ These are some of the hardest ECGs to interpret, where missed OMIs can delay reperfusion and false positives can trigger unnecessary cath lab activation.
- 🫀 Physicians missed 41% of true OMIs, while the Queen of Hearts AI missed only 11%, suggesting meaningful improvement in detection.
- 🎯 The AI also reduced overcalling of non-OMI ECGs, which could help decrease unnecessary catheterization laboratory activations.
📝 Introduction
Rapid identification of occlusive myocardial infarction (OMI) is essential to ensure timely catheterization laboratory activation (CLA). [1,2] Traditional STEMI criteria rely on measurement-based ST-segment elevation thresholds, but many acute coronary occlusions occur without classic ST elevation. [3,4] These presentations, often referred to as STEMI-equivalents, are among the most challenging ECGs to interpret, and failure to recognize them can delay reperfusion therapy and worsen patient outcomes. [4]
Conversely, several ECG findings can mimic STEMI despite the absence of true coronary occlusion, resulting in false-positive catheterization laboratory activations in 9% to 23% of cases. [5,6] These STEMI-mimics may demonstrate ST-segment elevation meeting traditional millimeter-based criteria but are caused by conditions such as left ventricular hypertrophy, early repolarization, bundle branch blocks, old myocardial infarction, or Brugada pattern. [4,7] In addition, conditions that cause myocardial injury without coronary occlusion, including takotsubo cardiomyopathy, myocarditis, and post-cardiac arrest changes, can create further diagnostic uncertainty. [7] Misinterpretation of these patterns may lead to unnecessary catheterization, exposing patients to contrast and radiation, increasing costs, and delaying diagnosis and treatment of the actual underlying condition.
Artificial intelligence tools for ECG interpretation have recently been developed to assist clinicians in identifying patterns associated with acute coronary occlusion. The Queen of Hearts (QoH) AI system, a deep neural network trained on millions of ECGs, analyzes 12-lead ECG tracings to identify patterns consistent with OMI, including cases that may not meet traditional STEMI criteria, while also helping differentiate true occlusion from common STEMI mimics.
This study evaluated whether the QoH AI algorithm could outperform emergency physicians and cardiologists in interpreting some of the most challenging ECGs in acute coronary care—STEMI-equivalents and STEMI-mimics—to determine whether cath lab activation was warranted.
🧾 Paper
Shroyer S, Mehta S, Thukral N, Smiley K, Mercaldo N, Meyers HP, Smith SW. Accuracy of cath lab activation decisions for STEMI-equivalent and mimic ECGs: physicians vs AI (Queen of Hearts by PMcardio). Am J Emerg Med.2025;97:193-199. PMID: 40763602
⚙️ What They Did
📈 Results
Metric | Emergency Physicians | Cardiologists | Queen of Hearts AI |
|---|---|---|---|
True Positive | 282 | 194 | 8 |
False Positive | 133 | 77 | 1 |
False Negative | 195 | 184 | 1 |
True Negative | 344 | 301 | 8 |
Total ECG Interpretations | 954 | 756 | 18 |
Sensitivity | 59% (95% CI 36–79) | 51% (95% CI 31–71) | 89% (95% CI 76–95) |
Specificity | 72% (95% CI 57–83) | 80% (95% CI 67–88) | 89% (95% CI 78–95) |
Accuracy | 65.6% (95% CI 51–78) | 65.5% (95% CI 51–77) | 88.9% (95% CI 82–93) |
Potential Unnecessary CLA | 32% of caths | 28% of caths | 11% of caths |
Potential Missed OMIs | 41% of OMIs | 49% of OMIs | 11% of OMIs |
- 95 total physicians participated: 53 emergency physicians and 42 cardiologists
- The cardiologist group included 23 general cardiologists, 15 interventional cardiologists, and 4 electrophysiologists
- Median clinical experience was 7 years among emergency physicians, with IQR 3.0–15
- Median clinical experience was 15 years among cardiologists, with IQR 9.2–21
- Participants were recruited from a San Antonio, Texas hospital system with two community hospitals and four standalone EDs
💥 Critical Results
- QoH AI had higher overall accuracy than physicians (89% vs 66%; p < 0.001).
- Emergency physicians and cardiologists performed similarly (65.6% [95% CI 51–78] vs 65.5% [95% CI 51–77]; p = 0.97).
- QoH AI outperformed both groups in sensitivity and specificity, achieving 88.9% sensitivity and 88.9% specificity.
- Emergency physicians identified 59% of OMIs with 72% specificity, while cardiologists identified 51% of OMIswith 80% specificity.
- For difficult ECG subtypes, hyperacute T-wave OMI was more often recognized by EPs than cardiologists (57% vs 24%; p = 0.002), whereas RBBB with fascicular block was more often recognized by cardiologists (64% vs 42%; p = 0.030).
- QoH AI correctly classified 3 of 4 ECG patterns that more than half of physicians misclassified, though it missed mSgarbossa-positive LBBB and LV aneurysm.
💪 Strengths
- Blinding: Physicians and AI were blinded to outcomes and OMI prevalence and interpreted ECGs without clinical context, ensuring decisions were based solely on ECG interpretation.
- Reference Standard: Diagnostic accuracy was compared against a reference standard incorporating angiography, troponins, echocardiography, and clinical follow-up.
- Focus on High-Risk ECGs: Evaluation of complex STEMI-equivalent and STEMI-mimic patterns prone to diagnostic error
- Inter-Specialty Comparison: head-to-head comparison of performance between emergency physicians and cardiologists
- Experienced Participants: Participants were residency-trained physicians with substantial experience (median 7 years for emergency physicians and 15 years for cardiologists), increasing clinical relevance.
⚠️ Limitations
- Case Selection Bias: ECGs were a small set of hand-selected examples of STEMI-equivalents and STEMI-mimics rather than randomly sampled cases, likely exaggerating diagnostic difficulty.
- Artificial Study Environment: The controlled survey environment does not reflect real-world ED conditions such as time pressure, interruptions, and competing clinical priorities. In practice, these pressures could further worsen physician performance, potentially widening the gap between physician and AI accuracy.
- Spectrum Bias: The study used a simulated 50% OMI prevalence, which differs from real-world ED chest pain populations where OMI prevalence is much lower.
- Misclassification: One ECG was mischaracterized as modified Sgarbossa-positive LBBB, although sensitivity analysis showed minimal impact on overall study conclusions.
- Undefined Troponin Threshold: “Significantly elevated troponin” was not defined for cases with TIMI 3 flow, which could introduce some misclassification of OMI status. However, this limitation is likely minor since all participants interpreted the same ECG set
🗣️ Discussion
Not All Occlusions Meet STEMI Criteria
The traditional STEMI vs NSTEMI framework is increasingly being challenged by the OMI vs NOMI paradigm, which focuses on identifying acute coronary occlusion rather than relying solely on ST-segment elevation thresholds. Multiple studies have demonstrated that approximately one-third of patients classified as NSTEMI have a completely occluded coronary artery at angiography, and these patients may experience mortality rates comparable to those with classic STEMI. [8-10] This suggests there is a substantial subset of patients who could benefit from earlier cath lab activation despite not meeting traditional STEMI criteria. Recognizing these subtle ECG patterns requires careful scrutiny and expertise, and the findings of this study suggest that AI-assisted ECG interpretation may help clinicians identify these occlusion patterns that are frequently missed by physicians alone.
Different Specialties, Different Priorities
One interesting finding from this study is that emergency physicians and cardiologists performed almost identically overall, with diagnostic accuracy of roughly 66% for both groups. However, their error patterns differed in ways that reflect their clinical roles. Emergency physicians demonstrated higher sensitivity for identifying STEMI-equivalent ECGs, whereas cardiologists demonstrated higher specificity for STEMI mimics. This likely reflects differing priorities: emergency physicians are generally more focused on not missing an occlusion, while cardiologists are more focused on avoiding unnecessary cath lab activations. The high rate of cath lab activation for the Wellens case by both groups further illustrates evolving clinical thinking, with most physicians recognizing Wellens syndrome as a STEMI-equivalent that warrants urgent intervention rather than delayed evaluation.
ECG Patterns That Still Challenge Experienced Clinicians
The study also highlights specific ECG patterns that remain challenging even for experienced clinicians. Hyperacute T-wave OMI and RBBB with fascicular block generated the greatest disagreement between emergency physicians and cardiologists, suggesting these patterns are particularly prone to diagnostic uncertainty and inter-specialty debate. Importantly, this was not a group of novice interpreters—participants were experienced physicians with a median of 7 years of practice among emergency physicians and 15 years among cardiologists—making the diagnostic difficulty of these ECG patterns even more notable. Interestingly, physicians performed well in identifying several classic STEMI mimics such as LVH and benign early repolarization, but accuracy dropped substantially when bundle branch blockswere present. These findings suggest that conduction abnormalities combined with ischemic patterns may represent one of the important ongoing educational gaps in ECG interpretation.
Potential Clinical Role for AI
Perhaps the most striking finding was the magnitude of difference between physicians and the AI model. Physicians missed 41% of true OMIs and overcalled 32% of non-OMIs, whereas the Queen of Hearts AI missed only 11% and overcalled 11%. While this study was conducted in a controlled environment with a small number of selected ECGs, the results suggest a potential clinical impact if these findings hold in real-world practice. Improved detection of subtle OMI patterns could reduce missed occlusions while simultaneously decreasing unnecessary cath lab activations. In theory, this could translate to better patient outcomes and more efficient use of cardiology resources. Rather than replacing physician interpretation, tools like QoH may ultimately function best as decision support systems, helping clinicians recognize complex ECG patterns that are difficult even for experienced physicians to consistently identify.
State of the Evidence
The Queen of Hearts AI model has now been evaluated in at least six recent peer-reviewed studies, including large international validation cohorts, real-world cath lab activation registries, and physician–AI comparison studies.
Study (Year) | Design | Dataset / Patients (n) | Study Groups / Comparison | Primary Outcome | Key Findings |
Multicenter international validation | ~18,000 ECGs | QoH AI vs STEMI criteria vs ECG experts | AI accuracy 90.9%, sensitivity 80.6%, specificity 93.7% for OMI detection | AI significantly improved detection of occlusion MI compared to use of STEMI criteria | |
Retrospective STEMI activation cohort | 304 STEMI pathway activations | QoH AI vs STEMI criteria | Sensitivity 89.2% vs 68.3%; specificity 72.9% vs 51.7%; accuracy 82.9% vs 61.8% | AI correctly identified 86/118 (73%) of false-positive cath lab activations | |
Retrospective case series | 42 OMI patients (78 ECGs) | QoH AI vs conventional ECG algorithms | AI detected OMI in 72% of first ECGs labeled “normal” by computer interpretation | Demonstrates ability of AI to identify occlusions missed by conventional ECG algorithms as “normal” | |
Cross-sectional physician vs AI study | 95 physicians interpreting 18 ECGs | Emergency physicians vs cardiologists vs QoH AI | AI accuracy 88.9% vs physicians 65.6% (p<0.001) | Missed OMI 11% vs 41%; unnecessary cath activations 11% vs 32% | |
Retrospective substudy of DOMI-ARIGATO cohort | 808 patients with ACS; 53 acute LAD TIMI-0 occlusions analyzed | QoH AI vs STEMI criteria vs expert interpretation | STEMI criteria missed 38% of acute LAD occlusions on the first ECG | Both AI and expert ECG interpretation detected 100% of LAD occlusions on the first ECG in cases missed by STEMI criteria (20 cases) | |
Multicenter registry validation study | Multicenter cohort of 1032 emergent CCL activations | Retrospective QoH AI analysis applied to real-world STEMI activation casesa | Sensitivity 92% vs 71% for detecting true STEMI/OMI | AI markedly reduced false-positive STEMI activations by reclassifying 277 of 306 (91%) biomarker-negative FPAs correctly. |
📘 Author's Conclusion
“QoH AI significantly outperformed physicians, suggesting a potential to inform cath lab activation processes that improve patient care and cardiology resource utilization.”
💬 Our Conclusion
In this study, the QoH AI system demonstrated higher diagnostic accuracy than both emergency physicians and cardiologists when interpreting challenging STEMI-equivalent and STEMI-mimic ECG patterns. This is not the first study demonstrating robust performance of this AI model, and taken together the available evidence suggests that AI-assisted ECG interpretation may be ready for use as an adjunct to physician interpretation, helping clinicians recognize subtle occlusion patterns that are frequently difficult to detect.
🚨 Clinical Bottom Line
Once FDA approved, the Queen of Hearts will be ready for clinical use as tool to support physicians in ECG interpretation.
👉 FAQ
- What is the Queen of Hearts AI ECG model?
The Queen of Hearts (QoH) is an AI ECG interpretation tool designed to identify patterns consistent with occlusive myocardial infarction (OMI), including cases that may not meet traditional STEMI criteria. - Can AI outperform physicians on STEMI-equivalent and STEMI-mimic ECGs?
In this study, the QoH AI model demonstrated higher overall diagnostic accuracy than both emergency physicians and cardiologists when interpreting challenging STEMI-equivalent and STEMI-mimic ECGs. - What are common STEMI-equivalent ECG patterns?
Common STEMI-equivalent patterns include posterior MI, de Winter T waves, hyperacute T waves, modified Sgarbossa-positive left bundle branch block or paced rhythm, and other subtle findings that suggest acute coronary occlusion despite the absence of classic STEMI criteria. - What are common causes of STEMI-mimic ECG patterns?
Common STEMI mimics include left ventricular hypertrophy, early repolarization, bundle branch blocks, old myocardial infarction with persistent ST elevation, and Brugada pattern. Other conditions such as takotsubo cardiomyopathy and myocarditis can also create similar ECG appearances. - Can AI improve cath lab activation decisions?
AI may help clinicians better recognize subtle occlusion patterns and distinguish true OMI from mimics, which could reduce both missed OMIs and unnecessary cath lab activations when used as an adjunct to physician interpretation.
📚 References
- Bhatt DL, Lopes RD, Harrington RA. Diagnosis and Treatment of Acute Coronary Syndromes: A Review. JAMA. 2022;327(7):662–675.
- Rao SV, O’Donoghue ML, Ruel M, et al. 2025 ACC/AHA/ACEP/NAEMSP/SCAI Guideline for the Management of Patients With Acute Coronary Syndromes: A Report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines. J Am Coll Cardiol. 2025;85(22):2135-2237.
- Writing Committee, Kontos MC, de Lemos JA, et al. 2022 ACC Expert Consensus Decision Pathway on the Evaluation and Disposition of Acute Chest Pain in the Emergency Department: A Report of the American College of Cardiology Solution Set Oversight Committee. J Am Coll Cardiol. 2022;80(20):1925-1960.
- Kola M, Shuka N, Meyers HP, Zaimi Petrela E, Smith SW. OMI/NOMI: Time for a New Classification of Acute Myocardial Infarction. J Clin Med. 2024;13(17):5201. Published 2024 Sep 2.
- Agrawal A, Lu M, Kanjanahattakij N, et al. ECG clues for false ST-segment elevation myocardial infarction activations. Coron Artery Dis. 2019;30(6):406-412.
- Larson DM, Menssen KM, Sharkey SW, et al. “False-Positive” Cardiac Catheterization Laboratory Activation Among Patients With Suspected ST-Segment Elevation Myocardial Infarction. JAMA. 2007;298(23):2754–2760.
- Herman R, Mumma BE, Hoyne JD, et al. AI-Enabled ECG Analysis Improves Diagnostic Accuracy and Reduces False STEMI Activations: A Multicenter U.S. Registry. JACC Cardiovasc Interv. 2026;19(2):145-156.
- Ricci F, Martini C, Scordo DM, et al. ECG Patterns of Occlusion Myocardial Infarction: A Narrative Review. Ann Emerg Med. 2025;85(4):330-340.
- Hung CS, Chen YH, Huang CC, et al. Prevalence and outcome of patients with non-ST segment elevation myocardial infarction with occluded “culprit” artery – a systemic review and meta-analysis. Crit Care. 2018;22(1):34. Published 2018 Feb 9.
- Khan AR, Golwala H, Tripathi A, et al. Impact of total occlusion of culprit artery in acute non-ST elevation myocardial infarction: a systematic review and meta-analysis. Eur Heart J. 2017;38(41):3082-3089.
Post Peer Reviewed By: Anand Swaminathan, MD (Twitter/X: @EMSwami), Propersi, DO (X:@Marco_propersi)
👤 Guest Authors
Joseph Bove, DO FAAEM
Anika Suri, MD
🔎 Your Deep-Dive Starts Here
MDCalc Wars: HEART Score Vs. EDACS
Chest pain is one of the most common—and anxiety-inducing—presentations in ...
A Winning Hand in Cardiology: Queen of Hearts AI Model Enhances OMI Detection
Can AI spot occlusive MI on a single 12-lead ECG ...
REBEL Core Cast 113.0 – ACS Therapies and Management
Take Home Points: All STEMIs should be loaded with dual ...
REBEL Cast Ep114: High Flow O2, Suspected ACS, and Mortality?
Background: Historically, we have treated acute coronary syndrome with supplemental ...
Retrospective Validation of High-Sensitivity Troponin with 5 Clinical Decision Instruments
Background: Chest pain is the second most common reason for presentation ...
Rebellion21: Time to Evolve – Redefining Coronary Ischemia – The OMI/NOMI Paradigm via Tarlan Hedayati, MD
In this 14-minute presentation from Rebellion in EM 2021, Dr. ...
