BACKGROUND
Atrial fibrillation (AF) is one of the most common and relevant cardiac arrhythmias in terms of morbidity and mortality. Early and accurate diagnosis is essential for adequate medical attention and the reduction of serious complications. Artificial intelligence (AI) has emerged as a technological tool for the detection of AF with the automated analysis of the electrocardiogram (ECG).
OBJECTIVE
The objective of this exploratory systematic review was to identify the AI models applied to the multi-lead ECG for the detection of AF, to describe the accuracy and generalization capacity of the models.
METHODS
The search strategy was applied in 2 bibliographic databases, Scopus on February 5, 2024, and PubMed on February 6, 2024. Of 23,785 studies, 23 were selected following strict inclusion and exclusion criteria and underwent a rigorous analysis, data synthesis and quality assessment.
RESULTS
The exposition of the characteristics was carried out in two groups according to the number of derivations of the input ECG of the algorithm. In the 2-lead group, the average sensitivity was 0.95, specificity 0.94 and accuracy 0.94. In the 12-lead test, the average sensitivity was 0.91, specificity 0.96 and accuracy 0.97.
CONCLUSIONS
Despite a promising accuracy, the generalizability was limited by the methodological quality of the model development process and the need for clinical validation of the algorithms with real-world data.
CLINICALTRIAL
Not applicable