Key Takeaways:
- A deep learning ECG model significantly outperforms clinician interpretation in rapidly identifying patients who require coronary revascularization and accurately diagnosing type 1 myocardial infarction, performing comparably to high-sensitivity cardiac troponin assays.
- The AI-based tool offers rapid risk stratification in emergency departments, potentially reducing treatment delays and improving clinical outcomes, especially for patients without obvious ST-segment elevations on initial ECGs.
A new artificial intelligence (AI)-based electrocardiogram (ECG) model developed for rapid risk stratification in emergency departments accurately identifies patients who require urgent coronary revascularization, outperforming clinician interpretation and performing comparably to high-sensitivity cardiac troponin (hs-TnT) testing. Results from the international study, presented at the American College of Cardiology’s Annual Scientific Session (ACC.25) and simultaneously published in the European Heart Journal, highlight the potential of AI to expedite care for patients presenting with possible acute coronary syndrome (ACS).
The study included 199,359 emergency department visits across two independent cohorts: 180,686 visits from the U.S. (training and internal testing) and 18,673 from Europe (external validation). The deep learning ECG model was trained exclusively on raw 12-lead ECG waveforms without additional clinical data. In the internal test cohort, the AI model achieved an area under the receiver operating characteristic curve (AUROC) of 0.91 (95% CI, 0.91–0.91), significantly outperforming human clinicians (AUROC, 0.65; 95% CI, 0.54–0.76) and conventional cardiac troponin T (AUROC, 0.71). In the external validation cohort, the model demonstrated an AUROC of 0.81 for predicting coronary revascularization and 0.85 for detecting type 1 myocardial infarction, outperforming clinician ECG interpretation (AUROC, 0.70 and 0.74, respectively) and approaching the performance of hs-TnT (AUROC, 0.85 for revascularization; 0.87 for type 1 MI). The model classified patients into low-, intermediate-, and high-risk groups, providing rapid and actionable triage recommendations.
Dr. Antonius Büscher, lead author and clinician-scientist at University Hospital Münster, highlighted the clinical significance: “Whereas patients with ST segment elevation have changes in the ECG that are quite distinct, there is a large number of patients who need urgent coronary revascularization who do not have ST segment elevation; because the alterations on the ECG are not as clear-cut, those patients can experience treatment delays. Our goal was to accelerate this process to identify patients who might need revascularization earlier.”
Explainability analyses indicated that the AI model primarily utilized clinically established ECG markers, particularly focusing on ST-segment features and T-wave abnormalities, providing transparency and boosting clinical confidence. Researchers noted the model’s high diagnostic accuracy remained robust across diverse patient populations, emphasizing its potential broad applicability. The authors emphasized that the AI model is not meant to replace clinical judgement but rather to enhance diagnostic capabilities, potentially alerting medical staff or triggering cardiac troponin testing in patients with an MI but not initially diagnosed.
A prospective study is planned to evaluate the AI model’s impact on clinical practice and patient outcomes.