Related Experiment Video
Updated: Mar 27, 2026
08:10
Estimating Bilateral Atrial Function by Cardiovascular Magnetic Resonance Feature Tracking in Patients with Paroxysmal Atrial Fibrillation
Published on: July 20, 2022
2.2K
A two-stage deep learning framework for predicting the onset of Atrial fibrillation using RR interval-based
Yongbin Lee1, Yeonsik Noh2,3, Allan Walkey4
1Department of Biomedical Engineering, University of Connecticut, Storrs, CT 06269, USA.
Biocybernetics and Biomedical Engineering
|March 25, 2026
Summary
This study introduces a deep learning model to predict atrial fibrillation (AF) onset one hour in advance using RR intervals. The framework offers high accuracy for early detection in critically ill patients.
Area of Science:
- Cardiology
- Artificial Intelligence
- Medical Informatics
Background:
- Atrial fibrillation (AF) is a common arrhythmia with significant stroke and heart failure risks.
- Current AF detection methods are accurate but AF onset prediction is underexplored.
- Paroxysmal AF often goes undetected, increasing mortality risk, especially in intensive care unit (ICU) patients.
Purpose of the Study:
- To develop and validate a deep learning framework for predicting AF onset up to one hour in advance.
- To address the critical need for early AF prediction in hospitalized patients, particularly in ICUs.
- To utilize RR intervals (RRIs) for robust AF onset prediction.
Main Methods:
- A two-stage deep learning framework was designed, combining convolutional and bidirectional long short-term memory (BiLSTM) networks.
- The first stage extracts features from RRIs, and the second stage predicts AF onset using another BiLSTM and a classifier.
- The model was evaluated using subject-wise testing and external independent dataset validation.
Main Results:
- Subject-wise testing yielded high performance metrics, including AUROC of 0.980.
- External validation demonstrated strong predictive power with AUROC of 0.976 and AUPRC of 0.966.
- The model achieved sensitivity of 0.848 and specificity of 0.978 on the independent dataset.
Conclusions:
- The proposed two-stage deep learning framework provides state-of-the-art AF onset prediction.
- The model offers lightweight computational complexity and flexible training, with potential for clinical interpretation via masking techniques.
- This robust framework enables prediction of AF up to one hour in advance, allowing for timely preventive interventions.
