Diagnostic accuracy of an AI-based ECG algorithm for atrial flutter: a prospective comparison with electrophysiological study

Introduction Despite advances in automated AF detection, current algorithms struggle to identify typical atrial flutter (AFL) due to its regular ventricular response and characteristic 2:1 atrioventricular conduction. As a result AFL is often misclassified or labeled inconclusive, even in clearly readable tracings, creating diagnostic uncertainty and increasing the burden of manual ECG review. These limitations highlight the need for validated AI tools capable of accurately detecting AFL. Aim To evaluate the diagnostic accuracy of a novel AI-based ECG algorithm for detecting typical atrial flutter using electrophysiological study as the reference standard. Methods and Results In this single-center diagnostic study, 600 patients referred for AFL ablation were enrolled. Standard 12-lead ECGs were obtained prior to EPS and analyzed by both the AI algorithm and two independent electrophysiologists. EPS confirmed typical AFL in 281 patients (47%), sinus rhythm in 252 (42%), and atrial fibrillation in 67 (11%). For detection of typical AFL, the AI algorithm achieved an accuracy of 97% (95% confidence interval [CI] 95–98), sensitivity 97%, specificity 98%, positive predictive value 98%, and negative predictive value 97%, demonstrating non-inferiority compared with manual expert interpretation. When applied to lead I only, simulating single-lead consumer devices, accuracy was 90% (CI 86–93), while manual interpretation by cardiologists reached 98% accuracy (κ = 0.95). Conclusion A novel AI-based algorithm enables highly accurate detection of typical AFL from 12-lead ECGs, with diagnostic performance comparable to expert cardiologists and confirmed by EPS. Integration of such algorithms into smart devices and clinical ECG platforms could enhance rhythm classification and support earlier recognition of AFL in ambulatory care.