Related Experiment Video
Updated: Jul 3, 2026
05:03
Patient Directed Recording of a Bipolar Three-Lead Electrocardiogram using a Smartwatch with ECG Function
Published on: December 11, 2019
Deep Neural Networks for Automatic Atrial Fibrillation Detection Using Long-Term Ambulatory Electrocardiography:
Jagdeep Sedha1,2, Jukka A Lipponen3, Antti Aho1
1School of Medicine, Faculty of Health Sciences, University of Eastern Finland, Yliopistonranta 8, Kuopio, 70210, Finland, 358 443358523.
JMIR Cardio
|June 30, 2026
Summary
A deep neural network (DNN) model effectively detects atrial fibrillation (AF) and atrial flutter (AFL) in long-term ECGs. This AI tool shows high accuracy, aiding early diagnosis and reducing clinician workload for these common arrhythmias.
Area of Science:
- Cardiology
- Artificial Intelligence
- Medical Diagnostics
Background:
- Atrial fibrillation (AF) and atrial flutter (AFL) are prevalent cardiac arrhythmias affecting millions globally.
- Early detection is critical due to stroke risk, yet diagnosis is challenging due to asymptomatic and episodic presentations.
- Traditional electrocardiogram (ECG) interpretation demands expert input and struggles with poor-quality recordings.
Purpose of the Study:
- To evaluate a deep neural network (DNN) model for detecting AF/AFL in diverse, long-term ambulatory ECG data.
- To assess the DNN model's real-world clinical performance and generalizability.
- To determine the model's utility in supporting automated screening and reducing manual review burdens.
Main Methods:
- Developed a DNN model using a large dataset combining public, prior study, and hospital-specific long-term ECG recordings (over 15,000 patients).
- Validated the model's accuracy and generalizability on a separate test set of 1010 patients with expert-annotated long-term ECGs.
- Assessed performance across varied patient demographics, comorbidities, coexisting arrhythmias, and ECG qualities.
Main Results:
- The DNN model achieved high sensitivity (96.4%) and specificity (>99.99%) for time-level AF/AFL detection.
- At the recording level, sensitivity was 100% and specificity 98.77%, with a low false positive rate (1.2%).
- The model demonstrated consistent high performance across diverse patient characteristics and ECG qualities.
Conclusions:
- The developed DNN model shows significant potential for automated screening of AF and AFL in long-term ambulatory ECGs.
- This AI-driven approach can effectively support clinical practice by reducing the manual workload associated with ECG interpretation.
- The model's robust performance across varied data suggests its applicability in real-world clinical settings for improved arrhythmia detection.
Keywords:
AIECG analysisartificial intelligenceatrial fibrillationcardiac arrhythmia detectiondeep neural networkselectrocardiogramelectrocardiogram analysismachine learning
