Derivation of an Artificial Intelligence - based Electrocardiographic Model for the Detection of Acute Coronary Occlusive Myocardial Infarction (ACOMI)
Braiana Ángeles Díaz Herrera1, Mariana Robles Ledesma1, Carlos Alan Castro García1, María del Pilar Espinosa Martínez1, Santiago March Mifsut1, Alexandra Arias Mendoza1, Diego Araiza Garaygordobil1.
¹Instituto Nacional de Cardiología, Mexico City, Mexico.
Abstract
Background
Current guidelines suggest to classify patients with acute coronary syndromes according to STEMI criteria, however, this classification has important limitations. Large series have documented that up to 15% of patients initially classified as NSTEMI will show evidence of total coronary occlusion on index angiography. Artificial intelligence (AI) holds promise in this regard, previous endeavors have shown that AI
EKG based algorithms can enhance diagnostic accuracy, considering OMI signatures are spatial in nature, representing a significant variation among patients.
Methods
This is a prospective, cross-sectional study based on the development of an AI-based ECG model capable of detecting an Acute Coronary Occlusion Myocardial Infarction (ACOMI). Single-standard 12-lead electrocardiograms from 366 patients who presented to the emergency department with an acute coronary syndrome between 2017 to 2023 at the National Institute of Cardiology, Mexico City, Mexico were contemplated. ECGs were also evaluated by two expert cardiologists independently and blinded for the clinical outcome, each was presented with the images and asked to determine whether the patient had an STEMI based on the criteria of the fourth Universal Definition of Myocardial Infarction and the suspicion of OMI based on the state-of-the-art criteria. ACOMI was angiographically defined as the presence of one of the following: total thrombotic occlusion, TIMI thrombus grade 2 or higher + TIMI grade flow 1 or less, or the presence of a subocclusive lesion. Patients were classified into four groups: ST elevation + OMI (Occlusion Myocardial Infarction), No ST elevation + OMI, ST elevation + NOMI (Non-occlusion Myocardial Infarction)and No ST elevation + NOMI.
Results
Patients were 62.95 ± 11.57 years old, 80.47% male, and the proportion of diabetes was 41.71%, hypertension and dyslipidemia were 52.71% and 28.45%, respectively.
The ACOMI AI model accomplished an AUC of 0.87 in identifying occlusion myocardial infarction as our primary outcome, compared to the AUC achieved by ECG experts, AUC: 0.53. Our model showed a higher sensitivity (83-86%) in identifying OMI compared to the STEMI criteria (70%) and ECG experts (80%).
Conclusion
This study aimed to assess the performance of an AI-EKG based algorithm capable of detecting ACOMI in the setting of patients with non ST-segment elevation myocardial infarction (or NSTEMI). In the present study including patients with ACS, an AI-EKG based algorithm outperformed EKG experts and demonstrated a higher diagnostic precision for the detection of ACOMI. Further research is needed in order to externally validate our algorithm and understand the role of AI-based models in the care of patients with ACS.