Specificity of the Occlusion Myocardial Infarction-Trained PMcardio "Queen of Hearts" AI-ECG in Inflammatory Myopericardial Syndromes — A Comparison With Guideline Criteria and Large Language Models

Overview

Inflammatory myopericardial syndromes can mimic occlusion myocardial infarction (OMI) on ECG, risking unnecessary catheterization lab activations. This retrospective study found the OMI-trained PMcardio Queen of Hearts (QoH) AI-ECG model reached 95.0% specificity for excluding OMI in 242 patients with these syndromes, versus 78.1% for ESC STEMI criteria and 26.9%–70.2% for three general-purpose LLMs. Discordant cases showed diffuse, nonterritorial ECG patterns typical of inflammatory disease rather than focal ischemia.

Published in: JACC Advances
Published on: 28 July 2026

Background

Inflammatory myopericardial syndromes (IMPS) — pericarditis, myocarditis, and overlap phenotypes — are common non-ischemic causes of acute chest pain whose ECG changes can mimic occlusion myocardial infarction (OMI). AI-ECG models trained on angiographic OMI endpoints, notably PMcardio's Queen of Hearts (QoH), have outperformed STEMI criteria and expert interpretation, but whether QoH avoids false-positive OMI calls on inflammatory ST changes remained untested. This study assessed QoH specificity in an IMPS-only cohort against ESC STEMI criteria, a 4-variable formula, and three general-purpose large language models (LLMs).

Methods

The authors conducted a single-center retrospective study of consecutive emergency department patients with adjudicated IMPS at the University & Hospital of Fribourg (2010–2025). Diagnoses were set by two cardiologists blinded to QoH/LLM outputs, with obstructive coronary disease excluded by angiography and/or cardiac MRI. Each patient's first 12-lead ECG was analyzed with QoH (PMcardio App v3.4.1) and compared against ESC STEMI criteria, the 4-variable formula, and three LLMs (ChatGPT 5.2, Claude Opus 4.5, Gemini 3). The primary endpoint was ECG-level specificity for non-OMI classification, with secondary analyses by ECG territory and IMPS phenotype.

Results

Among 242 patients with confirmed IMPS, QoH reached 95.0% specificity (95% CI: 91.5%-97.1%), versus 78.1% for ESC STEMI criteria and 72.9%-73.3% for the 4-variable formula. All three LLMs performed worse: Claude Opus 4.5 was best at 70.2%, followed by ChatGPT at 62.4% and Gemini at 26.9% (all P < 0.001 vs QoH). Specificity was higher in pure (myo)pericarditis (97.7%) than in (peri)myocarditis with myocardial involvement (87.7%, P = 0.004). Among 43 discordant ESC-positive/QoH-negative cases, review found recurrent inflammatory features — notched J-points (74.4%), concave ST morphology (74.4%), multi-territorial STE (67.4%), and PR depression (30.2%) — rather than focal occlusion patterns.

Conclusion

In this IMPS-only cohort, the OMI-trained Queen of Hearts model showed substantially higher specificity for excluding OMI than ESC STEMI criteria, the 4-variable formula, and three general-purpose LLMs, supporting its potential to reduce unnecessary catheterization laboratory activations in inflammatory myopericardial disease.