Diagnostic accuracy of artificial intelligence (AI)-based models for acute myocardial infarction: a systematic review and meta-analysis

Abstract Background Early diagnosis of acute myocardial infarction (AMI) is crucial to reduce scar fibrosis and prevent further clinical complications related to left ventricular dysfunction or arrhythmias. The diagnosis of AMI is based on ECG, in addition to other non-invasive imaging techniques and clinical and laboratory parameters. Despite continuous improvements in diagnostic tools, manual interpretation of these methods can lead to potential inconsistencies between different observers. To address this issue, automated computer-based diagnostic systems using artificial intelligence (AI) have been developed. Several studies in the literature have presented machine learning (ML) or deep learning (DL) models to quickly and accurately assess the diagnosis of AMI in order to optimize treatment. Aim To evaluate the performance of integrating AI-based models into standard clinical practice for the diagnosis of AMI using data from all validation and test cohorts available in the literature. Methods A systematic search of PubMed, Scopus and Google Scholar identified studies evaluating the diagnostic accuracy of artificial intelligence algorithms for AMI diagnosis. Only data from validation and test cohorts were considered for the present analysis. Sensitivity and specificity were analyed for different types of AI (machine learning or deep learning) and for all MIs, STEMI only and NSTEMI only. Hierarchical models were used to estimate the summary receiver operating characteristic curve, sensitivity and specificity. Results A total of 23 studies with 198,955 patients and 27 AI-based diagnostic algorithms were included in the analysis. Our cohort of patients admitted with chest pain was predominantly male (62.9%) with a mean age of 62 years (IQR 59-64), and AMI was found in 90,088 (45.2%) cases. In most cases (14/27, 52%) ECG represented the primary data source of the model. The integration of an AI-based algorithm into standard practice was highly efficient in diagnosing AMI with an AUC of 0.97 (sensitivity 0.93, specificity 0.92). This result remained consistent when considering the ability to diagnose STEMI (AUC 0.96, sensitivity 0.89, specificity 0.93) or NSTEMI (AUC 0.97, sensitivity 0.95, specificity 0.90) separately. Finally, the use of a deep learning model was associated with slightly better performance (AUC 0.98, sensitivity 0.93, specificity 0.93) than machine learning (AUC 0.97, sensitivity 0.93, specificity 0.90). Conclusions The integration of AI diagnostic algorithms into standard practice is associated with particularly efficient performance, which could accelerate the detection of myocardial infarction and prevent complications due to delayed treatment. These results were consistent for both STEMI and NSTEMI considered separately. DL models showed slightly better accuracy than ML models.AUC of AI-based models for AMI diagnosis