Artificial intelligence capabilities in identifying atrial fibrillation using baseline sinus rhythm ECG : a...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Atrial fibrillation (AF) is a common arrhythmia with significant adverse outcomes.
• Paroxysmal AF detection is challenging due to inefficient methods and lack of unified screening.
• Artificial intelligence (AI) offers potential for early AF detection and intervention.

Purpose of the Study:

• To assess the effectiveness of AI in detecting AF using baseline sinus rhythm-ECGs (SR-ECGs).
• To identify factors influencing the performance of AI models in AF detection.

Main Methods:

• Systematic review of studies using AI to detect AF from baseline SR-ECGs.
• Double-screening of references up to May 2024 across eight databases.
• Quality assessment using the Quality Assessment of Diagnostic Accuracy Studies-2 tool.
• Performance metrics summarized using medians with subgroup analyses by AI type and AF confirmation timeframe.

Main Results:

• Analysis of 14 studies and 33 AI models, including 1,459,653 patients.
• Median accuracy was 58.0%, sensitivity 62.0%, specificity 57.8%, precision 52.0%, and AUC 0.740.
• Deep learning models showed superior sensitivity and AUC compared to traditional machine learning.
• Models with a 31-day confirmation window demonstrated higher accuracy and AUC than those with a 1-year window.

Conclusions:

• AI-enhanced SR-ECG shows potential for identifying AF patients.
• Deep learning models and a 31-day confirmation window are more effective for AF detection.
• Further research is required to evaluate clinical benefits and cost-effectiveness.