JMIR Publications

ABSTRACT

Background:

Occlusion myocardial infarction (OMI) on the ECG can present as STEMI, NSTEMI, ST depression, or other patterns (e.g., de Winter T-waves, hyperacute T waves), posing a diagnostic challenge. Even experienced physicians may overlook OMI on ECG, leading to delays in catheterization laboratory activation. Recently, smartphone applications utilizing artificial intelligence (AI) have been shown to exhibit a promising accuracy in ECG interpretation in large datasets.

Objective:

The objective of our study was to retrospectively compare the diagnostic performance of an AI-based application (PMCardio®) with that of interventional cardiologists in interpreting non-pathognomonic ECGs that generated debates about diagnosis.

Methods:

In total, 33 ECGs that were diagnostically challenging upon patient admission to the emergency department were included. A group of 3 interventional cardiologists, blinded to the final diagnosis, independently assessed whether the ECG suggested OMI requiring immediate invasive coronary angiography. The diagnostic performance of the cardiologists was compared to that of the AI application, using the results of coronary angiography as the gold standard and considering any lesion with ≥90% stenosis as positive.

Results:

Overall, 29 (88%) patients had a final diagnosis of cardiac origin for their chest pain, including 23 acute coronary syndromes, of which 13 (57%) were OMI. The 3 cardiologists achieved sensitivities of 62%,85%,62%, negative predictive values of 74%,83%,69%, specificities of 70%, 50%, 55%, and positive predictive values of 57%52% and 47%, respectively. Their overall misclassification rates were 33%, 36% and 42%. In contrast, AI analysis demonstrated a sensitivity and NPV of 100%, a specificity of 60%, a PPV of 62%, and an overall misclassification rate of 24%.

Conclusions:

In this retrospective study of challenging ECGs, an AI application demonstrated better diagnostic performance of OMI as compared to cardiologists, underscoring its reliability for clinical use. These findings highlight the potential of AI to enhance diagnostic accuracy and support clinical decision-making.