In September the FDA granted De Novo authorization to the Queen of Hearts AI-ECG model, the first built to flag acute coronary occlusion rather than classic STEMI criteria. A 24,511-patient Swedish study shows why that distinction matters, and what we still don't know about using it in the field.
Educational reading for clinicians. Follow your current protocols and local guidance.
STEMI Criteria Miss Occlusions
Every medic who has transmitted a 12-lead knows the frustration. The patient looks like an MI, the troponin is coming, and the ECG doesn't meet millimeter criteria. The cardiology world has spent years arguing that the question isn't ST elevation, it's is there an occluded artery. That's the occlusion MI (OMI) paradigm, and a recent review puts the share of OMIs missed by STEMI/NSTEMI triage at roughly 25 to 34%.
On September 3, 2026, the FDA granted De Novo authorization (DEN250044) to Powerful Medical's Queen of Hearts (ACS AI ECG Model), an algorithm trained to flag occlusion from a single 12-lead. The FDA also authorized a Predetermined Change Control Plan, which lets the company update the model within an agreed scope without a new submission.
What the Best Recent Data Shows
The most useful study for us came out in JACEP Open in October 2026. Swedish researchers ran the model, standard STEMI criteria and the Glasgow ECG algorithm on every chest-pain patient in an existing ED cohort:
- 24,511 patients, of whom 467 (1.9%) had an OMI confirmed by angiography and expert adjudication.
- Sensitivity: Queen of Hearts 52%, STEMI criteria 23%, Glasgow algorithm 32%.
- Specificity: 99% versus 98%.
- Positive predictive value: 51% versus 17%.
| Method | Sensitivity | Specificity | PPV |
|---|---|---|---|
| Queen of Hearts AI | 52% | 99% | 51% |
| STEMI criteria | 23% | 98% | 17% |
| Glasgow algorithm | 32% | 98% | 26% |
| Extended STEMI criteria | 41% | 95% | 14% |
Read that twice. The AI more than doubled the occlusions caught, and its positive calls were three times as likely to be right. It still missed about half of them. A normal AI read is not a rule-out, and the authors and reviewers say so: ruling out MI still takes serial high-sensitivity troponin.
What This Means on the Truck
It's decision support, not a cath lab key. The authorization creates a cleared tool. It doesn't change who activates the lab or what your protocol says.
The field population wasn't in this study. The Swedish cohort was ED patients, and it excluded those taken straight from the ambulance to the coronary care unit. Prehospital performance needs its own data.
We don't know the exact label yet. The public De Novo listing doesn't show the indications text, so I can't tell you whether prehospital 12-lead use is inside it. Before your agency buys anything, ask the vendor for the indications for use in writing.
It rewards good acquisition. An algorithm reading a 12-lead with loose limb leads and baseline wander is reading garbage. Lead placement and a clean tracing matter more, not less.
The direction is clear, though. The software is getting better at spotting occlusion than the criteria we were taught. The medic who understands why a borderline 12-lead worries them will use it best.
Sources
- FDA De Novo database — DEN250044, Queen of Hearts (ACS AI ECG Model)
- JACEP Open (2026) — Improved Detection of Acute Coronary Occlusion MI by an AI ECG Model in Swedish EDs (PubMed 42614578)
- Cardiovascular Revascularization Medicine (2026) — AI-enhanced ECG for ACS triage: a state-of-the-art review (PubMed 42364947)
Jonathan B, career paramedic
Published October 23, 2026