Deep learning model may help detect type 1 MI and revascularization need from ECG patterns, reveals research

A new study published in the European Heart Journal showed that IL-Using electrocardiogram (ECG) data, a new artificial intelligence (AI)-powered tool may identify type 1 myocardial infarction (MI) more accurately at a level comparable to that of a high-sensitivity troponin T (hs-TnT) assay.

Inconclusive electrocardiogram (ECG) or biomarker data can make it difficult to identify individuals with acute coronary syndrome who need coronary revascularization. Antonius Büscher and colleagues created a deep learning model to identify ECG patterns linked to the risk of revascularization in order to direct additional evaluation and lower diagnostic ambiguity.

A convolutional neural network model was evaluated using a different test cohort (n=35,995), trained on 1,44,691 ED visits from a US cohort (60±19 years; 53% female; 0.6% revascularization), and compared to cardiac troponin T (TnT) and clinician ECG interpretation.

The results of 18,673 hospitalizations from Europe (55±21 years; 49% female; 1.5% revascularization; 1% type 1 MI) were externally validated for revascularization and type 1 MI. Area under the receiver operating characteristic curve (AUROC) served as the main performance indicator.

The model outperformed both traditional cardiac TnT (AUROC 0.71) and physician ECG interpretation in the test group, achieving an AUROC of 0.91. ECG model AUROC was 0.81 for revascularization and 0.85 for type 1 MI in the external validation cohort, while it was 0.67 and 0.74 for clinician interpretation and 0.85 and 0.87 for high-sensitivity (hs)-TnT, respectively. When compared to hs-TnT, the ECG model showed a lower sensitivity but a greater specificity.

The algorithm performed better than traditional hs-TnT testing and physicians' interpretation of the ECG in determining which individuals in the test population needed revascularization. Similar outcomes were obtained in the external validation cohort, where the ECG-AI model outperformed doctors in terms of interpretation but fell short of hs-TnT's predictive capabilities.

Overall, nearly 2,00,000 ECGs from two worldwide cohorts of patients with NSTE ACS who presented to the emergency department (ED) were used to train the deep-learning model. The findings imply that doctors may eventually get AI assistance when transferring suitable patients from the ED to the cath lab.

Reference:

Büscher, A., Plagwitz, L., Yildirim, K., Brix, T. J., Neuhaus, P., Bickmann, L., Menke, A. F., van Almsick, V. F., Pavenstädt, H., Kümpers, P., Heider, D., Varghese, J., & Eckardt, L. (2025). Deep learning electrocardiogram model for risk stratification of coronary revascularization need in the emergency department. European Heart Journal. https://doi.org/10.1093/eurheartj/ehaf254