Artificial intelligence-enabled electrocardiography algorithm for detection of occlusive myocardial infarction in suspected acute coronary syndromes without ST-segment elevation in clinical practice

Abstract Background Accurate recognition of occlusive myocardial infarction in acute coronary syndromes (ACS) without ST-elevation remains challenging, even for skilled clinicians. The use of artificial intelligence (AI) applied to ECG (AI-ECG) interpretation could improve accuracy of diagnosis and optimize patient management timing. Aim To assess the performance of AI-ECG in detecting occlusive myocardial infarction (OMI) based solely on automated ECG interpretation in comparison to human interpretation. Methods in this single-center prospective study, patients with symptoms suggestive of ACS without ST-elevation were included and treated according to ESC guidelines. All first ECGs at presentation were analyzed by clinicians. When NSTEMI was confirmed, coronary angiography was performed according to current ESC guidelines. Simultaneously, the same ECGs were also analyzed using a smartphone-based, commercially available, CE-certified AI-ECG algorithm and classified either as signs of occlusive myocardial infarction (AI-OMI) or AI-Rule-Out. Final diagnoses were used to assess the performance of both clinicians and AI-ECG. Results Among 1490 enrolled patients, AI-ECG classified 108 patients as AI-OMI (7%) and 1382 patients as AI-Rule-Out. According to the guideline-directed pathway, 1,207 were ruled out based on negative or stable serial troponin values while a total of 283 patients underwent coronary angiography to confirm the diagnosis. AI-ECG achieved 77% sensitivity and 99% specificity, with 98% negative predictive value and 84% positive predictive value for OMI, whereas guideline-directed therapy led to 88% specificity and 42% positive predictive value. There were 27 false negatives (2%) and 17 false positives (1%). Within this cohort of clinically suspected ACS patients, the AI algorithm outperformed human ECG interpretation by correctly identifying 84% of OMI cases. Conclusions The AI-ECG algorithm demonstrated superior accuracy in identifying and excluding occlusive myocardial infarction in patients with suspected ACS in comparison to conventional expert interpretation — prior to troponin. Its integration into routine practice may accelerate the diagnosis, giving access to an earlier treatment and promoting more efficient utilization of medical resources.