Artificial intelligence (AI) based analysis of electrocardiograms (ECG) can better detect severe heart attacks in emergency settings than human analysis, according to a new multicenter US registry study. The AI technology — PMcardio’s Queen of Hearts model — had a 4-fold reduction in false ST-segment elevation myocardial infarction (STEMI) activations, and its ability remained constant even with unconventional symptoms and atypical ECG patterns. It is one of the first large, real-world evaluations of an AI-based ECG model for STEMI triage in the emergency setting. The registry findings were presented Tuesday in a late-breaking clinical science session at the Transcatheter Cardiovascular Therapeutics (TCT) 2025 conference in San Francisco by the study’s senior author Timothy D. Henry, MD, from the Christ Hospital, Cincinnati. The study was published simultaneously in JACC: Cardiovascular Interventions. The results “Indicate that AI-enhanced STEMI diagnosis at the first medical contact has the potential to shorten time to treatment and reduce false activations,” said Dr. Henry in an accompanying American College of Cardiology press release. “This technology may be especially valuable in optimizing the transfer of STEMI patients from non-PCI centers to ensure timely and appropriate care.” Driver for AI The research was sparked by continued delays in achieving guideline-recommended reperfusion times despite advancements in regional STEMI systems and prehospital catheterization laboratory (CCL) activation, “especially at non-PCI centers and rural sites,” Dr. Henry and colleagues wrote. “Time is muscle,” they stressed, noting in the press release that time to reperfusion longer than 90 minutes raises the risk of mortality 3-fold for both STEMI and STEMI equivalents. Conversely, quick activation of the CCL is known to reduce delays in primary PCI intervention and improve clinical outcomes. Standardized protocols and pre-arranged transfer pathways have been put in place to improve primary PCI access, but these are often based on ECG interpretation by non-cardiologist clinicians. While this has improved PCI for those who need it, these rapid activation strategies also risk false-positive activation (FPA) rates, which are reported at 15% to 30%, thus leading to “staff fatigue, unnecessary resource utilization and potential patient harm,” the researchers said. The study A total 1,032 patients aged 18 years or older with suspected STEMI who triggered emergency cardiac CCL activation across 3 geographically diverse percutaneous coronary intervention (PCI) centers in the U.S. from January 2020 to May 2024 were included in the analysis. Sites included the Beth Israel Deaconess Medical Center in Boston, University of California-Davis Medical Center in Sacramento and Memorial Hermann-Texas Medical Center in Houston. Participants were identified from prospective, electronic, consecutive STEMI CCL activation logs reported to the National Cardiovascular Data Registry (NCDR) Chest Pain–Myocardial Infarction (MI) and CathPCI Registry. Patients had undergone standard index ECG triage and were then put through blinded retrospective AI ECG analysis with the Queen of Hearts system. The AI was trained to detect acute coronary occlusion and benign mimics. The reference standard was an angiographically confirmed culprit lesion with positive enzymes. Of the full cohort, 601 (58.2%) had confirmed STEMI through angiography and biomarkers. The rest were positives. Overall mean age of the 1,032 was 63.1 years — which remained similar between true STEMI and “mimics” (62.7 vs. 63.6 years; p = 0.299) — men made up the majority of patients (70.6%), and FPAs were more often walk-ins (18.3% of true STEMI vs 25.3% of mimics) than emergency medical services (EMS) arrivals (64.4% vs 58.7%, respectively). Mimics were significantly more likely to have higher heart rates ≥100 beats per minute (bpm, 24% of true STEMI vs 32.5% of mimics; P = 0.006), QRS duration ≥120 ms (11.8% vs 25.1%, respectively; P≤0.001), heart-rate-corrected QT interval (QTc) duration ≥440 ms (18.3% vs 33.9%, respectively; P≤0.001) and left bundle branch block (LBBB, 3.7% vs 14.2%, respectively; P≤0.001). In comparison to standard triage, the AI ECG model outperformed at diagnoses with a higher index ECG sensitivity detecting 553 out of 601 (92%; 95% confidence interval [CI]: 89.7 to 94.1) confirmed cases compared with 427 (71%; 95% CI: 67.4 to 74.6) in human analysis. It also delivered a 4-fold reduction in FPA rates, which were seen for 34 out of the 431 FPAs due to STEMI mimics for the AI analysis (7.9%; 95% CI: 6.4 to 9.6) compared with 180 in human analysis (41.8%; 95% CI: 38.9–44.7). Specificity was also superior (p<0.001). Dr. Henry and colleagues also highlighted the AI technology’s skill at identifying real vs mimic STEMI across clinically challenging subgroups, including atrial fibrillation, bundle branch block and STEMI equivalents. The model maintained consistent performance, with an area under the curve of 0.94 (95% CI: 0.92 to 0.95). The AI also reclassified 277 out of 306 (91%) of biomarker-negative FPAs correctly, the researchers said. The results support integration of AI-based ECG analysis into acute chest pain pathways, they said. “AI-driven ECG interpretation can bring the best of both worlds — identify true heart attacks early while reducing unnecessary activations,” said Robert Herman, MD, PhD, lead author of the study and a cardiovascular researcher at AZORG Hospital in Aalst, Belgium, in the accompanying TCT news release. “Improving the accuracy of triage at the first medical contact can streamline emergency care, reduce fatigue and strain on clinical teams, and ensure that patients who truly need urgent intervention receive it without delay.” Proceed with caution In an accompanying editorial, Mohamad Alkhouli, MD, MBA, cardiologist from the Mayo Clinic in Rochester, Minnesota, added that the researchers should be “commended for developing an operational AI model aimed at addressing one of the most complex and error-prone aspects of interventional cardiology practice — STEMI activation.” However, he warned that the AI model should be “interpreted with caution” since it was originally developed to detect occluded arteries, rather than STEMI. Both Dr. Alkhouli and the study’s authors called for calling for prospective implementation studies in the setting. In particular, Dr. Alkhouli said these should include diverse populations. “The true challenge is not proof of accuracy alone, but readiness—to integrate, regulate, and interpret AI as a complement to human judgment, particularly in high-stakes, time-sensitive clinical settings,” Dr. Alkhouli concluded. Source: 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 2025; DOI: 10.1016/j.jcin.2025.10.018. Image Caption: Timothy D. Henry, MD, presents during a press conference Tuesday, October 28, at the Transcatheter Cardiovascular Therapeutics (TCT) 2025 conference in San Francisco. Image Credit: Screenshot by Bailey G. Salimes, CRTonline.org.