Deep Learning-Enhanced ECG Analysis for the Prediction of Occult Atrial Fibrillation: A Multi-Center Prospective Validation Study

Original Research | 2025 | Volume 1 | Issue 1 | Page 50-59

Dr. Sayama Prasad, Professor, Cardiology, GMC-Asam

Dr. Arun Sharma, Professor, Cardiology, GMC-Jalgaow

Abstract

Deep Learning-Enhanced ECG Analysis for the Prediction of Occult Atrial Fibrillation: A Mu.

Download • 527KB

Background: Atrial Fibrillation (AFib) remains a leading cause of cryptogenic stroke due to its often paroxysmal and asymptomatic ("occult") nature. Traditional screening relies on symptomatic presentation or prolonged Holter monitoring, which often fails to capture intermittent episodes. This study evaluates a Deep Learning (DL) algorithm’s ability to identify a "digital signature" of AFib from a standard 10-second, 12-lead Electrocardiogram (ECG) recorded during normal sinus rhythm.

Methods: In this multi-center prospective validation study (2024–2026), we enrolled 15,000 patients across eight tertiary cardiac centers. All patients had a baseline "Normal Sinus Rhythm" ECG and underwent subsequent 14-day continuous patch monitoring to establish the ground truth. A Convolutional Neural Network (CNN), previously trained on over 1.2 million ECGs, was used to analyze the baseline strips for subtle micro-architectural changes in the P-wave and PR-segment.

Results: The DL-ECG model identified patients with occult AFib with an Area Under the Curve (AUC) of 0.91 (95% CI, 0.89–0.93). The algorithm demonstrated a Sensitivity of 84% and a Specificity of 88%. Notably, in patients flagged as "High Risk" by the AI who did not show AFib on initial 14-day monitoring, a 1-year follow-up revealed a 3.5x higher incidence of new-onset AFib compared to the "Low Risk" group ($p < 0.001$).

Discussion: These results suggest that AI can detect "structural and electrical remodeling" that precedes the clinical manifestation of AFib. This "Predictive Window" allows for a shift in stroke prevention strategy, identifying candidates for intensified monitoring or early anticoagulation. The study concludes that DL-enhanced ECG analysis serves as a powerful, low-cost screening tool, effectively turning a standard 10-second ECG into a long-term predictive biomarker for stroke risk.