Automated diagnosis of atrial fibrillation in 24-hour Holter recording based on deep learning:a study with randomized and real-world data validation

Peng Zhang, Fan Lin, Fei Ma, Yuting Chen, Dao Wen Wang, Xiaoyun Yang, Qiang LimedRxiv 2021 · DOI

Abstract

Atrial fibrillation (AF) screening via 24-hour Holter monitoring is labor-intensive for clinicians. This study presents a deep-learning-based AI model designed to automatically detect AF episodes and identify AF patients using RR intervals. Validated on a massive real-world dataset of nearly 20,000 recordings, the model achieved high sensitivity and specificity, processing a full day's data in seconds. This approach offers a scalable, cost-effective solution for clinical AF screening by reducing diagnostic workload and improving stroke prevention.

Introduction

Atrial fibrillation is a common heart rhythm disorder that significantly increases the risk of stroke and heart failure. While persistent AF is easily spotted, paroxysmal (intermittent) AF is often elusive. The gold standard for detection is the 24-hour Holter monitor, which records roughly 100,000 heartbeats per session. Manually reviewing this volume of data is a grueling, subjective task for cardiologists. While AI has shown promise, most previous models were trained on small, curated datasets of short ECG strips. This study bridges the gap by applying deep learning to large-scale, real-world clinical practice.

Methods and Model Architecture

The researchers moved away from traditional raw ECG waveform analysis, opting instead for RR intervals—the time elapsed between successive heartbeats. This approach requires 50x less storage space and is more robust against signal interference that often obscures P-wave morphology.

Overview of the deep learning pipeline for AF detection and patient identification.

The model utilizes a CLDNN (Convolutional, Long Short-Term Memory, and Deep Neural Network) architecture:

  1. CNN layers extract local features from sequences of 90 RR intervals.
  2. LSTM layers capture the temporal rhythm dependencies over time.
  3. Fully connected layers perform the final classification into AF or non-AF (NAF).

A critical clinical bridge was established: a patient is diagnosed with AF if the model detects at least one episode lasting 6 minutes or longer. This threshold is clinically significant as it is linked to a much higher risk of thromboembolism.

Datasets and Validation

The study is notable for its massive scale and "real-world" validation. It utilized a randomized clinical cohort of 3,000 recordings for training and initial testing, followed by a real-world clinical cohort of 19,979 consecutive recordings from a hospital center. This latter group represents the true variety of clinical patients, including those with pacemakers and other complex arrhythmias.

Flowchart illustrating the selection process for the randomized and real-world clinical datasets.

Key Results

The model demonstrated exceptional performance across both test environments:

  • Episode Detection (Real-World): Sensitivity of 0.971 and specificity of 0.997.
  • Patient Identification (Real-World): Sensitivity of 0.995 and specificity of 0.974.
  • Efficiency: The model processed 24 hours of data in approximately 4 seconds, compared to minutes for waveform-based AI.
  • Robustness: High specificity was maintained even in the presence of other irregular rhythms like premature contractions (PACs/VPCs).

Discussion and Limitations

The RR-interval-based model performed as well as or better than raw-ECG methods, proving highly robust against noise. However, some limitations remain. The model had slightly lower accuracy (0.854) in distinguishing AF from Atrial Tachycardia due to rhythm similarities. Additionally, "regularized" AF—cases where AF co-exists with a complete heart block or a pacemaker—can lead to false negatives as the rhythm becomes regular.

Despite these hurdles, the study demonstrates that deep learning can successfully automate AF identification in long-term recordings. This technology can be embedded into Holter analysis systems to assist clinical decision-making, improve screening efficiency, and reduce the overall cost of AF diagnosis.

Summary updated

Abstract

Summary Background With the increasing demand for atrial fibrillation (AF) screening, clinicians spend a significant amount of time in identifying the AF signals from massive electrocardiogram (ECG) data in long-term dynamic ECG monitoring. In this study, we aim to reduce clinicians’ workload and promote AF screening by using artificial intelligence (AI) to automatically detect AF episodes and identify AF patients in 24 h Holter recording. Methods We used a total of 22 979 Holter recordings (24 h) from 22 757 adult patients and established accurate annotations for AF by cardiologists. First, a randomized clinical cohort of 3 000 recordings (1 500 AF and 1 500 non-AF) from 3000 patients recorded between April 2012 and May 2020 was collected and randomly divided into training, validation and test sets (10:1:4). Then, a deep-learning-based AI model was developed to automatically detect AF episode using RR intervals and was tested with the test set. Based on AF episode detection results, AF patients were automatically identified by using a criterion of at least one AF episode of 6 min or longer. Finally, the clinical effectiveness of the model was verified with an independent real-world test set including 19 979 recordings (1 006 AF and 18 973 non-AF) from 19 757 consecutive patients recorded between June 2020 and January 2021. Findings Our model achieved high performance for AF episode detection in both test sets (sensitivity: 0.992 and 0.972; specificity: 0.997 and 0.997, respectively). It also achieved high performance for AF patient identification in both test sets (sensitivity:0.993 and 0.994; specificity: 0.990 and 0.973, respectively). Moreover, it obtained superior and consistent performance in an external public database. Interpretation Our AI model can automatically identify AF in long-term ECG recording with high accuracy. This cost-effective strategy may promote AF screening by improving diagnostic effectiveness and reducing clinical workload. Research in context Evidence before this study We searched Google Scholar and PubMed for research articles on artificial intelligence-based diagnosis of atrial fibrillation (AF) published in English between Jan 1, 2016 and Aug 1, 2021, using the search terms “deep learning” OR “deep neural network” OR “machine learning” OR “artificial intelligence” AND “atrial fibrillation”. We found that most of the previous deep learning models in AF detection were trained and validated on benchmark datasets (such as the PhysioNet database, the Massachusetts Institute of Technology Beth Israel Hospital AF database or Long-Term AF database), in which there were less than 100 patients or the recordings contained only short ECG segments (30-60s). Our search did not identify any articles that explored deep neural networks for AF detection in large real-world dataset of 24 h Holter recording, nor did we find articles that can automatically identify patients with AF in 24 h Holter recording. Added value of this study First, long-term Holter monitoring is the main method of AF screening, however, most previous studies of automatic AF detection mainly tested on short ECG recordings. This work focused on 24 h Holter recording data and achieved high accuracy in detecting AF episodes. Second, AF episodes detection did not automatically transform to AF patient identification in 24 h Holter recording, since at present, there is no well-recognized criterion for automatically identifying AF patient. Therefore, we established a criterion to identify AF patients by use of at least one AF episode of 6 min or longer, as this condition led to significantly increased risk of thromboembolism. Using this criterion, our method identified AF patients with high accuracy. Finally, and more importantly, our model was trained on a randomized clinical dataset and tested on an independent real-world clinical dataset to show great potential in clinical application. We did not exclude rare or special cases in the real-world dataset so as not to inflate our AF detection performance. To the best of our knowledge, this is the first study to automatically identifies both AF episodes and AF patients in 24 h Holter recording of large real-world clinical dataset. Implications of all the available evidence Our deep learning model automatically identified AF patient with high accuracy in 24 h Holter recording and was verified in real-world data, therefore, it can be embedded into the Holter analysis system and deployed at the clinical level to assist the decision making of Holter analysis system and clinicians. This approach can help improve the efficiency of AF screening and reduce the cost for AF diagnosis. In addition, our RR-interval-based model achieved comparable or better performance than the raw-ECG-based method, and can be widely applied to medical devices that can collect heartbeat information, including not only the multi-lead and single-lead Holter devices, but also other wearable devices that can reliably measure the heartbeat signals.
BibTeX
@misc{zhang2021automated,
  author = {Peng Zhang and Fan Lin and Fei Ma and Yuting Chen and Dao Wen Wang and Xiaoyun Yang and Qiang Li},
  title = {{Automated diagnosis of atrial fibrillation in 24-hour Holter recording based on deep learning:a study with randomized and real-world data validation}},
  year = {2021},
  doi = {10.1101/2021.08.25.21262591},
  howpublished = {medRxiv},
}
Figure