Detection Of Fibrillation Research Articles

Residual-attention deep learning model for atrial fibrillation detection from Holter recordings

Residual-attention deep learning model for atrial fibrillation detection from Holter recordings

Learning From Alarms: A Robust Learning Approach for Accurate Photoplethysmography-Based Atrial Fibrillation Detection Using Eight Million Samples Labeled With Imprecise Arrhythmia Alarms.

Atrial fibrillation (AF) is a common cardiac arrhythmia with serious health consequences if not detected and treated early. Detecting AF using wearable devices with photoplethysmography (PPG) sensors and deep neural networks has demonstrated some success using proprietary algorithms in commercial solutions. However, to improve continuous AF detection in ambulatory settings towards a population-wide screening use case, we face several challenges, one of which is the lack of large-scale labeled training data. To address this challenge, we propose to leverage AF alarms from bedside patient monitors to label concurrent PPG signals, resulting in the largest PPG-AF dataset so far (8.5 M 30-second records from 24,100 patients) and demonstrating a practical approach to build large labeled PPG datasets. Furthermore, we recognize that the AF labels thus obtained contain errors because of false AF alarms generated from imperfect built-in algorithms from bedside monitors. Dealing with label noise with unknown distribution characteristics in this case requires advanced algorithms. We, therefore, introduce and open-source a novel loss design, the cluster membership consistency (CMC) loss, to mitigate label errors. By comparing CMC with state-of-the-art methods selected from a noisy label competition, we demonstrate its superiority in handling label noise in PPG data, resilience to poor-quality signals, and computational efficiency.

Diagnostic accuracy for detecting atrial fibrillation using a novel machine learning algorithm in a blood pressure monitor

BackgroundEarly detection of atrial fibrillation (AF) is key for preventing strokes. Blood pressure monitors (BPMs) with built-in AF screening features have the potential for early detection at home. Recently, 2 BPMs (HEM-7371T1-AZ and HEM-7372T1-AZAZ, Omron Healthcare Co., Ltd.) that share a novel AF screening feature have been developed. Their AF screening feature utilizes an algorithm that incorporates machine learning, with the potential to improve diagnostic accuracy. ObjectiveThe purpose of this study was to evaluate the performance of this AF screening feature in a multicenter, prospective clinical study at 5 sites in the United States. MethodsA total of 559 subjects were enrolled for this study: 267 in AF cohort and 292 in the non-AF cohort. AF screening was performed in all subjects by the 2 Omron BPMs and by 1 Microlife BPM (BP 3MX1-3, WatchBP Home A, Microlife Corp.), and a simultaneous 12-lead electrocardiogram (ECG) was recorded for comparison. All 12-lead ECGs were interpreted by a board-certified cardiologist who was blinded to the BPM results. Sensitivity, specificity, and accuracy for the diagnosis of AF were calculated. ResultsOmron HEM-7371T1-AZ BPM had sensitivity of 95.1% (95% confidence interval [CI] 91.8%–97.4%), specificity 98.6% (95% CI 96.6%–99.7%), and accuracy of 97.0% (95% CI 95.2%–98.2%). Equivalent results were obtained with the Omron HEM-7371T1-AZAZ BPM. This compared favorably to the Microlife BPM (sensitivity 78.5%, 95% CI 73.1%–83.3%; specificity 97.6%, 95% CI 95.1%–99.0%; accuracy 88.4%, 95% CI 85.5%–91.0%). ConclusionThese data support both home and professional use of these novel Omron BPMs for the detection of AF.

Rate of atrial fibrillation by Holter-Stroke Risk Analysis in undetermined TIA/rapidly improving stroke symptoms patients.

Holter-SRA (Stroke Risk Analysis) is an automated analysis of ECG monitoring for Atrial Fibrillation (AF) detection. The aim of this study was to evaluate the rate of AF in undetermined TIA/Rapidly improving stroke symptoms (RISS) patients. Prospective study of undetermined TIA/RISS patients who presented to the emergency department. Early vascular studies (angio CT, transthoracic echocardiography and ECG) were performed in emergency department. The Holter-SRA device was placed for 2 h and the patients were classified into: confirmed AF, high risk of AF or low risk of AF. Prolonged ambulatory monitoring (7 days) was carried out every month for patients with a high-risk pattern. The results were evaluated until definitive detection of AF or low-risk pattern. The endpoints were rate of AF and vascular recurrence at 90 days. Over a period of 24 months, 83 undetermined TIA/RISS patients were enrolled. The mean age was 70 ± 10 years and 61% were men. The median ABCD2 score was 4 points (1-7). After 2 h of monitoring in the emergency department, AF was detected in one patient (1.2%), 51 patients with a low-risk pattern and 31 patients (37.3%) showed a high-risk pattern of AF. During the ambulatory monitoring, of the 31 patients high risk pattern patients, AF was diagnosed to 17 cases and of the 51 patients with a low-risk pattern, one case experienced a recurrent vascular due to undetected AF (1.9% false negative). Three patients (3.6%) suffered a vascular recurrence within the first 90 days, before AF diagnosis. In our study, AF was detected in 22.9% of the 83 patients with indeterminate TIA/RISS. Holter-SRA has allowed us to increase the detection of AF, especially those patients with a high-risk pattern in the first 3 months.

Automated interpretations of single-lead electrocardiograms predict incident atrial fibrillation: The VITAL-AF trial

BackgroundSingle-lead electrocardiograms (1L ECGs) are increasingly used for atrial fibrillation (AF) detection. Automated 1L ECG interpretation may have prognostic value for future AF in cases in which screening does not result in a short-term AF diagnosis. ObjectiveWe sought to investigate the association between automated 1L ECG interpretation and incident AF. MethodsVITAL-AF was a randomized controlled trial investigating the effectiveness of screening for AF by 1L ECGs. For this study, participants were divided into 4 groups based on automated classification of 1L ECGs. Patients with prevalent AF were excluded. Associations between groups and incident AF were assessed by Cox proportional hazards models adjusted for risk factors. The start of follow-up was defined as 60 days after the latest 1L ECG (as some individuals had numerous screening 1L ECGs). ResultsThe study sample included never screened (n = 16,306), normal (n = 10,914), other (n = 2675), and possible AF (n = 561). Possible AF had the highest AF incidence (5.91 per 100 person-years; 95% confidence interval [CI], 4.24–8.23). Possible AF was associated with greater hazard of incident AF compared with normal (adjusted hazard ratio, 2.48; 95% CI, 1.66–3.71). Other was associated with greater hazard of incident AF compared with normal (1.41; 95% CI, 1.04–1.90). ConclusionIn patients undergoing AF screening with 1L ECGs without prevalent AF or AF within 60 days of screening, presumptive positive and indeterminate 1L ECG interpretations were associated with future AF. Abnormal 1L ECG recordings may identify individuals at higher risk for future AF.

Wearable health monitoring: wave of the future or waste of time?

Atrial fibrillation is responsible for over 400,000 hospitalizations in the United States (US) each year. This costs the US health system over 4 billion each year. New smartwatches can constantly monitor pulse, oxygen saturation, and even heart rhythm. The FDA has provided clearance for select smartwatches to detect arrhythmias, including atrial fibrillation. These devices are not currently widely implemented as diagnostic tools. In this review, we delve into the mechanism of how smartwatches work as healthcare tools and how they capture health data. Additionally, we analyze the reliability of the data collected by smartwatches and the accuracy of their sensors in monitoring health parameters. Moreover, we explore the accessibility of smartwatches as healthcare tools and their potential to promote self-care among individuals. Finally, we assess the outcomes of using smartwatches in healthcare, including the limited studies on the clinical effects and barriers to uptake by the community. Although smartwatches are accurate for the detection of atrial fibrillation, they still face many hurdles, including access to aging populations and trust in the medical community.

Hemodynamic Bedside Monitoring Instrument with Pressure and Optical Sensors: Validation and Modality Comparison.

Results from two independent clinical validation studies for measuring hemodynamics at the patient's bedside using a compact finger probe are reported. Technology comprises a barometric pressure sensor, and in one implementation, additionally, an optical sensor for photoplethysmography (PPG) is developed, which can be used to measure blood pressure and analyze rhythm, including the continuous detection of atrial fibrillation. The capabilities of the technology are shown in several form factors, including a miniaturized version resembling a common pulse oximeter to which the technology could be integrated in. Several main results are presented: i) the miniature finger probe meets the accuracy requirements of non-invasive blood pressure instrument validation standard, ii) atrial fibrillation can be detected during the blood pressure measurement and in a continuous recording, iii) a unique comparison between optical and pressure sensing mechanisms is provided, which shows that the origin of both modalities can be explained using a pressure-volume model and that recordings are close to identical between the sensors. The benefits and limitations of both modalities in hemodynamic monitoring are furtherdiscussed.

0293 Evaluation of a Single Point of Contact Chest-worn HSAT with ECG-based Atrial Fibrillation Detection

0293 Evaluation of a Single Point of Contact Chest-worn HSAT with ECG-based Atrial Fibrillation Detection

Heart failure, recurrent vascular events and death in patients with ischemic stroke—results of the MonDAFIS study

Heart failure (HF) is associated with poor outcome after stroke, but data from large prospective trials are sparse.We assessed the impact of HF on clinical endpoints in patients hospitalized with acute ischemic stroke or transient ischemic attack (TIA) enrolled in the prospective, multicenter Systematic Monitoring for Detection of Atrial Fibrillation in Patients with Acute Ischemic Stroke (MonDAFIS) trial. HF was defined as left ventricular ejection fraction (LVEF) < 55% or a history of HF on admission. The composite of recurrent stroke, major bleeding, myocardial infarction, and all-cause death, and its components during the subsequent 24 months were assessed. We used estimated hazard ratios in confounder-adjusted models. Overall, 410/2562 (16.0%) stroke patients fulfilled the HF criteria (i.e. 381 [14.9%] with LVEF < 55% and 29 [1.9%] based on medical history). Patients with HF had more often diabetes, coronary and peripheral arterial disease and presented with more severe strokes on admission. HF at baseline correlated with myocardial infarction (HR 2.21; 95% CI 1.02–4.79), and all-cause death (HR 1.67; 95% CI 1.12–2.50), but not with major bleed (HR 1.93; 95% CI 0.73–5.06) or recurrent stroke/TIA (HR 1.08; 95% CI 0.75–1.57). The data were adjusted for age, stroke severity, cardiovascular risk factors, and randomization. Patients with ischemic stroke or TIA and comorbid HF have a higher risk of myocardial infarction and death compared with non-HF patients whereas the risk of recurrent stroke or major hemorrhage was similar. Trial registration number Clinicaltrials.gov NCT02204267.

Design of the Dutch multicentre study on opportunistic screening of geriatric patients for atrial fibrillation using asmartphone PPG app: the Dutch-GERAF study.

Screening of high-risk patients is advocated to achieve early detection and treatment of clinical atrial fibrillation (AF). The Dutch-GERAF study will address two major issues. Firstly, the effectiveness and feasibility of an opportunistic screening strategy for clinical AF will be assessed in frail older patients and, secondly, observational data will be gathered regarding the efficacy and safety of oral anticoagulation (OAC). This is amulticentre study on opportunistic screening of geriatric patients for clinical AF using asmartphone photoplethysmography (PPG) application. Inclusion criteria are age ≥ 65years and the ability to perform at least three PPG recordings within 6months. Exclusion criteria are the presence of acardiac implantable device, advanced dementia or asevere tremor. The PPG application records patients' pulse at their fingertip and determines the likelihood of clinical AF. If clinical AF is suspected after apositive PPG recording, aconfirmatory electrocardiogram is performed. Patients undergo acomprehensive geriatric assessment and afrailty index is calculated. Risk scores for major bleeding (MB) are applied. Standard laboratory testing and additional laboratory analyses are performed to determine the ABC-bleeding risk score. Follow-up data will be collected at 6months, 12months and 3years on the incidence of AF, MB, hospitalisation, stroke, progression of cognitive disorders and mortality. The Dutch-GERAF study will focus on frail older patients, who are underrepresented in randomised clinical trials. It will provide insight into the effectiveness of screening for clinical AF and the efficacy and safety of OAC in this high-risk population. NCT05337202.

Edge AI Model Deployed for Real-Time Detection of Atrial Fibrillation Risk during Sinus Rhythm.

Objectives: The study aimed to develop a deep learning-based edge AI model deployed on electrocardiograph (ECG) devices for the real-time detection of atrial fibrillation (AF) risk during sinus rhythm (SR) using standard 10 s, 12-lead electrocardiograms (ECGs). Methods: A novel approach was used to convert standard 12-lead ECGs into binary images for model input, and a lightweight convolutional neural network (CNN)-based model was trained using data collected by the Japan Agency for Medical and Research Development (AMED) between 2019 and 2022. Patients over 40 years old with digital, SR ECGs were retrospectively enrolled and divided into AF and non-AF groups. The data labeling was supervised by cardiologists. The dataset was randomly allocated into training, validation, and internal testing datasets. External testing was conducted on data collected from other hospitals. Results: The best-trained model achieved an AUC of 0.82 and 0.80, sensitivity of 79.5% and 72.3%, specificity of 77.8% and 77.7%, precision of 78.2% and 76.4%, and overall accuracy of 78.6% and 75.0% in the internal and external testing datasets, respectively. The deployed model and app package utilized 2.5 MB and 40 MB of the available ROM and RAM capacity on the edge ECG device, correspondingly. The processing time for AF risk detection was approximately 2 s. Conclusions: The model maintains comparable performance and improves its suitability for deployment on resource-constrained ECG devices, thereby expanding its potential impact to a wide range of healthcare settings. Its successful deployment enables real-time AF risk detection during SR, allowing for timely intervention to prevent AF-related serious consequences like stroke and premature death.

Diagnostic Performance of an Automated Blood Pressure Monitor With an Irregular Heartbeat Algorithm Designed to Detect Atrial Fibrillation.

Background: Early detection of atrial fibrillation (AF) remains an unsolved challenge and because the greatest risk factor for AF is hypertension, blood pressure (BP) monitors with AF detectors have been developed. We evaluated the clinical performance of an irregular heartbeat (IHB) algorithm built into an A&D automated BP monitor for AF diagnosis. Methods and Results: Each of the 239 enrolled patients underwent BP measurement 3 times using the A&D UM-212 with the IHB algorithm. Real-time 3-lead ECG was recorded using automated ECG analysis software. Independent of the ECG analysis software results, 2 cardiologists interpreted the ECG and made the final diagnosis. Of the 239 patients, 135 were in sinus rhythm, 31 had AF, and 73 had non-AF arrhythmias. The respective sensitivity, specificity, and accuracy of the IHB algorithm for AF diagnosis were 98.9%, 91.2%, and 92.2% for the per-measurement evaluation, and 96.8%, 95.7%, and 95.8% for the per-patient evaluation (3/3 positive measurements). The respective sensitivity, specificity, and accuracy of the ECG analysis software for AF diagnosis were 91.4%, 97.9%, and 97.1% for the per-measurement evaluation, and 77.4%, 99.5%, and 96.7% for the per-patient evaluation (3/3 positive measurements). Conclusions: The IHB algorithm built into an A&D automated BP monitor had high diagnostic performance for AF in general cardiology patients, especially when multiple measurements were obtained.

Prediction and Detection of Ventricular Fibrillation Using Complex Features and AI-Based Classification

We analyzed the possibility of detecting and predicting ventricular fibrillation (VF), a medical emergency that may put people’s lives at risk, as the medical prognosis depends on the time in which medical personnel intervene. Therefore, besides immediate detection of VF, the possibility of predicting VF 40 or even 50 min in advance was analyzed. For testing the proposed algorithm, we used ECG signals taken from the MIT-BIH database, namely, Malignant Ventricular Ectopy Database, Sudden Cardiac Death Holter Database and Normal Sinus Rhythm Database. The presented method is based on features extracted from the ECG signals in the time domain, frequency domain and complexity measures. For VF detection, the possibility of identifying the VF episode in the first 3 s after its occurrence was tested. For this, the first 3 s immediately after the appearance of VF were cut out and the features were computed on these sections. For VF prediction, 3 min of the ECG signal clipped 40 or 50 min before VF onset was used. Then, on these pieces of ECG signal, the specific features were calculated for 1 s segments. For the normal signal situation, 3 min was randomly selected from the database with normal ECGs. For the classification or detection stage, both an MLP-type neural network and the classifiers from the Machine Learning toolbox of the MATLAB® environment were used. The results obtained for both detection and classification are over 94% in both cases. The novelty of our results compared to those previously obtained is the time interval with which the possibility of prediction was analyzed, namely, 50 min in advance of the VF installation date. This means that the patient will be informed that it is possible to suffer a VF and has time to take the necessary measures to overcome a possible medical emergency.

Age-Dependent Detection of Atrial Fibrillation with Implantable Cardiac Monitors in Patients with Cryptogenic Stroke.

Background Continuous monitoring using implantable cardiac monitors (ICMs) results in atrial fibrillation (AF) detection rates of up to 30% in patients with cryptogenic stroke (CS). Although higher age is an independent risk factor for AF, there are no age-specific recommendations for the implantation of ICM. Objective The aim of this study was to analyze age-related AF rates in patients with CS and continuous rhythm monitoring, to determine the rates of oral anticoagulation (OAC) and recurrent cerebrovascular events (stroke or transient ischemic attack) in patients with ICM-detected AF, and to describe the temporal relationship of AF detection and recurrent cerebrovascular events. Methods In this observational study, patients with CS provided with ICMs were systematically followed. All patients underwent 72-hour electrocardiography monitoring, transcranial Doppler ultrasound, and transthoracic echocardiography prior to ICM insertion. Follow-up included a regular outpatient presentation every 3 months with medical history, physical examination, and interrogation of the ICM. Results One-hundred eighty-six patients (mean age: 65 ± 12 years, 54% female) were included in this analysis. AF was detected in 6, 27, 56, and 65% ( p < 0.001) of patients aged less than 60, 60 to 69, 70 to 79, and more than or equal to 80 years, respectively. All patients with AF under 60 years had an impaired left ventricular systolic function. OAC was initiated in 85% of the patients with AF. Recurrent cerebrovascular events occurred in 34 patients of whom 14 had a diagnosis of AF. In nine patients, AF was diagnosed before the occurrence of a recurrent cerebrovascular event. Conclusion AF prevalence increased with age and was absent in CS patients younger than 60 years and with preserved left ventricular ejection fraction. The temporal relationship of AF and recurrent cerebrovascular events was weak.

Higher Frequency of Premature Atrial Contractions Correlates With Atrial Fibrillation Detection after Cryptogenic Stroke.

Covert atrial fibrillation (AF) is a major cause of cryptogenic stroke. This study investigated whether a dose-dependent relationship exists between the frequency of premature atrial contractions (PACs) and AF detection in patients with cryptogenic stroke using an insertable cardiac monitor (ICM). We enrolled consecutive patients with cryptogenic stroke who underwent ICM implantation between October 2016 and September 2020 at 8 stroke centers in Japan. Patients were divided into 3 groups according to the PAC count on 24-hour Holter ECG: ≤200 (group L), >200 to ≤500 (group M), and >500 (group H). We defined a high AF burden as above the median of the cumulative duration of AF episodes during the entire monitoring period. We evaluated the association of the frequency of PACs with AF detection using log-rank trend test and Cox proportional hazard model and with high AF burden using logistic regression model, adjusting for age, sex, CHADS2 score. Of 417 patients, we analyzed 381 patients with Holter ECG and ICM data. The median age was 70 (interquartile range, 59.5-76.5), 246 patients (65%) were males, and the median duration of ICM recording was 605 days (interquartile range, 397-827 days). The rate of new AF detected by ICM was higher in groups with more frequent PAC (15.5%/y in group L [n=277] versus 44.0%/y in group M [n=42] versus 71.4%/y in group H [n=62]; log-rank trend P<0.01). Compared with group L, the adjusted hazard ratios for AF detection in groups M and H were 2.11 (95% CI, 1.24-3.58) and 3.23 (95% CI, 2.07-5.04), respectively, and the adjusted odds ratio for high AF burden in groups M and H were 2.57 (95% CI, 1.14-5.74) and 4.25 (2.14-8.47), respectively. The frequency of PACs was dose-dependently associated with AF detection in patients with cryptogenic stroke.

PRR30, pRR3.25% and Asymmetrical Entropy Descriptors in Atrial Fibrillation Detection.

Early detection of atrial fibrillation (AF) is essential to prevent stroke and other cardiac and embolic complications. We compared the diagnostic properties for AF detection of the percentage of successive RR interval differences greater than or equal to 30 ms or 3.25% of the previous RR interval (pRR30 and pRR3.25%, respectively), and asymmetric entropy descriptors of RR intervals. Previously, both pRR30 and pRR3.25% outperformed many other heart rate variability (HRV) parameters in distinguishing AF from sinus rhythm (SR) in 60 s electrocardiograms (ECGs). The 60 s segments with RR intervals were extracted from the publicly available Physionet Long-Term Atrial Fibrillation Database (84 recording, 24 h Holter ECG). There were 31,753 60 s segments of AF and 32,073 60 s segments of SR. The diagnostic properties of all parameters were analysed with receiver operator curve analysis, a confusion matrix and logistic regression. The best model with pRR30, pRR3.25% and total entropic features (H) had the largest area under the curve (AUC)-0.98 compared to 0.959 for pRR30-and 0.972 for pRR3.25%. However, the differences in AUC between pRR30 and pRR3.25% alone and the combined model were negligible from a practical point of view. Moreover, combining pRR30 and pRR3.25% with H significantly increased the number of false-negative cases by more than threefold. Asymmetric entropy has some potential in differentiating AF from SR in the 60 s RR interval time series, but the addition of these parameters does not seem to make a relevant difference compared to pRR30 and especially pRR3.25%.

Exploring the power of photoplethysmogram matrix for atrial fibrillation detection with integrated explainability

Atrial Fibrillation (AF) detection is paramount for cardiovascular health due to its potential complications. In this study, we investigate the utility of Photoplethysmogram (PPG) for continuous heart rate monitoring and detection of patients with AF. Our approach centres on creating PhotoplethysmoMatrices (PPMs) and leverages Explainable Artificial Intelligence (XAI) techniques to enable accurate AF patient classification with interpretability.Our method involves transforming PPG data into multiple PPM images, which represent aligned peaks within the PPG signal, presented as heatmaps. The diagnostic architecture is a lightweight and efficient Convolutional Neural Network (CNN) combined with attention mechanisms for model transparency.Remarkably, our approach achieves a 100% classification accuracy across multiple experiments, even with variations in the number of users and training images. Furthermore, the attention module underscores the significance of peak positioning and shifting in AF patient detection.Overall, our research makes a substantial contribution to the field of AF patient classification using PPG signals. The combination of image-based preprocessing techniques and explainable architectures enhances accuracy and interpretability, promising improved diagnostic capabilities in clinical settings.

Region-Disentangled Diffusion Model for High-Fidelity PPG-to-ECG Translation

The high prevalence of cardiovascular diseases (CVDs) calls for accessible and cost-effective continuous cardiac monitoring tools. Despite Electrocardiography (ECG) being the gold standard, continuous monitoring remains a challenge, leading to the exploration of Photoplethysmography (PPG), a promising but more basic alternative available in consumer wearables. This notion has recently spurred interest in translating PPG to ECG signals. In this work, we introduce Region-Disentangled Diffusion Model (RDDM), a novel diffusion model designed to capture the complex temporal dynamics of ECG. Traditional Diffusion models like Denoising Diffusion Probabilistic Models (DDPM) face challenges in capturing such nuances due to the indiscriminate noise addition process across the entire signal. Our proposed RDDM overcomes such limitations by incorporating a novel forward process that selectively adds noise to specific regions of interest (ROI) such as QRS complex in ECG signals, and a reverse process that disentangles the denoising of ROI and non-ROI regions. Quantitative experiments demonstrate that RDDM can generate high-fidelity ECG from PPG in as few as 10 diffusion steps, making it highly effective and computationally efficient. Additionally, to rigorously validate the usefulness of the generated ECG signals, we introduce CardioBench, a comprehensive evaluation benchmark for a variety of cardiac-related tasks including heart rate and blood pressure estimation, stress classification, and the detection of atrial fibrillation and diabetes. Our thorough experiments show that RDDM achieves state-of-the-art performance on CardioBench. To the best of our knowledge, RDDM is the first diffusion model for cross-modal signal-to-signal translation in the bio-signal domain.

Optimization of Using Multiple Machine Learning Approaches in Atrial Fibrillation Detection Based on a Large-Scale Data Set of 12-Lead Electrocardiograms: Cross-Sectional Study.

Atrial fibrillation (AF) represents a hazardous cardiac arrhythmia that significantly elevates the risk of stroke and heart failure. Despite its severity, its diagnosis largely relies on the proficiency of health care professionals. At present, the real-time identification of paroxysmal AF is hindered by the lack of automated techniques. Consequently, a highly effective machine learning algorithm specifically designed for AF detection could offer substantial clinical benefits. We hypothesized that machine learning algorithms have the potential to identify and extract features of AF with a high degree of accuracy, given the intricate and distinctive patterns present in electrocardiogram (ECG) recordings of AF. This study aims to develop a clinically valuable machine learning algorithm that can accurately detect AF and compare different leads' performances of AF detection. We used 12-lead ECG recordings sourced from the 2020 PhysioNet Challenge data sets. The Welch method was used to extract power spectral features of the 12-lead ECGs within a frequency range of 0.083 to 24.92 Hz. Subsequently, various machine learning techniques were evaluated and optimized to classify sinus rhythm (SR) and AF based on these power spectral features. Furthermore, we compared the effects of different frequency subbands and different lead selections on machine learning performances. The light gradient boosting machine (LightGBM) was found to be the most effective in classifying AF and SR, achieving an average F1-score of 0.988 across all ECG leads. Among the frequency subbands, the 0.083 to 4.92 Hz range yielded the highest F1-score of 0.985. In interlead comparisons, aVR had the highest performance (F1=0.993), with minimal differences observed between leads. In conclusion, this study successfully used machine learning methodologies, particularly the LightGBM model, to differentiate SR and AF based on power spectral features derived from 12-lead ECGs. The performance marked by an average F1-score of 0.988 and minimal interlead variation underscores the potential of machine learning algorithms to bolster real-time AF detection. This advancement could significantly improve patient care in intensive care units as well as facilitate remote monitoring through wearable devices, ultimately enhancing clinical outcomes.

Lead-Specific Performance for Atrial Fibrillation Detection in Convolutional Neural Network Models Using Sinus Rhythm Electrocardiography.

Background: We developed a convolutional neural network (CNN) model to detect atrial fibrillation (AF) using the sinus rhythm ECG (SR-ECG). However, the diagnostic performance of the CNN model based on different ECG leads remains unclear. Methods and Results: In this retrospective analysis of a single-center, prospective cohort study, we identified 616 AF cases and 3,412 SR cases for the modeling dataset among new patients (n=19,170). The modeling dataset included SR-ECGs obtained within 31 days from AF-ECGs in AF cases and SR cases with follow-up ≥1,095 days. We evaluated the CNN model's performance for AF detection using 8-lead (I, II, and V1-6), single-lead, and double-lead ECGs through 5-fold cross-validation. The CNN model achieved an area under the curve (AUC) of 0.872 (95% confidence interval (CI): 0.856-0.888) and an odds ratio of 15.24 (95% CI: 12.42-18.72) for AF detection using the eight-lead ECG. Among the single-lead and double-lead ECGs, the double-lead ECG using leads I and V1 yielded an AUC of 0.871 (95% CI: 0.856-0.886) with an odds ratio of 14.34 (95% CI: 11.64-17.67). Conclusions: We assessed the performance of a CNN model for detecting AF using eight-lead, single-lead, and double-lead SR-ECGs. The model's performance with a double-lead (I, V1) ECG was comparable to that of the 8-lead ECG, suggesting its potential as an alternative for AF screening using SR-ECG.