Artificial Intelligence Software for Detecting Paroxysmal Atrial Fibrillation from Sinus Rhythm Monitor ECG:...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Early detection of paroxysmal atrial fibrillation (pAF) is crucial for timely intervention and stroke prevention.
• Deep learning algorithms applied to electrocardiograph (ECG) data offer potential for improved pAF detection.
• Holter monitoring provides valuable ECG data for analyzing cardiac arrhythmias.

Purpose of the Study:

• To develop and prospectively evaluate a deep-learning Holter ECG algorithm for detecting pAF from sinus rhythm.
• To assess the algorithm's patient-level performance in identifying individuals who develop pAF within a 7-day period.
• To validate the algorithm as a potential triage tool for intensified rhythm monitoring.

Main Methods:

• A deep-learning convolutional model was trained on 20,000 30-second sinus rhythm ECG segments.
• Tenfold cross-validation and a separate validation set were used to tune the model and select an operating threshold.
• A multicenter prospective study evaluated the algorithm on consecutive 30-second sinus rhythm blocks, with concurrent 7-day patch monitoring for outcome confirmation.

Main Results:

• The tuned algorithm achieved 84.9% sensitivity and 69.9% specificity on a validation set.
• In the prospective clinical trial, the algorithm demonstrated 91.7% sensitivity and 65.0% specificity for detecting pAF within 7 days.
• No device-related adverse events were reported during the study.

Conclusions:

• An AI algorithm analyzing short sinus rhythm Holter ECG segments can effectively identify patients who develop pAF within 7 days.
• The developed AI tool shows promise as a triage mechanism to guide intensified rhythm monitoring for pAF detection.
• This approach supports earlier intervention strategies for stroke prevention in patients at risk for pAF.