Use of artificial intelligence for detection and managing atrial fibrillation—narrative review of the current literature

Introduction

Atrial fibrillation (AF) is characterized by an abnormal heartbeat or arrhythmia that can cause clots in blood vessels associated with the heart and brain which can ultimately lead to myocardial infarction or stroke. It is estimated that AF can affect up to 5 million people in The United States (1). Stroke is one important and substantially devastating consequence that results in substantial morbidity and therefore mortality related to AF. According to one estimate, 15% of all stroke cases in the United States are attributed to AF (2). Due to the considerable morbidity and mortality related to AF early detection and prediction of AF is very important for early management.

AF is predicted based on the findings from electrocardiogram (ECG) as well as risk scores like Cohorts for Aging and Research in Genomic Epidemiology-AF (CHARGE-AF). The use of CHARGE-AF score for clinical prediction can be cumbersome as it requires the abstraction of data either manually or through some automated process (3). In the context of AF, the data comes from several resources that include but are not limited to data from the routine care of the patients, that include data from investigations and radiographs, and also from wearable devices and in some cases from remote patient monitoring systems. It is important to process this collected data and leverage it in a way to optimize the predictive ability as well as management of AF to reduce potential morbidity and mortality (4).

Recent advancements in artificial intelligence (AI) and machine learning (ML) have opened doors for using the data to develop models to early detect medical conditions. While AI and ML are terms that have been used broadly, these are actually two very distinct concepts, with ML being a subset of AI. More generally, AI can be described as the ability of machines to act just like human beings in terms of the intelligence and doing of work under realistic conditions, such as solving problems, decision-making, or language understanding. ML refers to the algorithms and techniques that enable systems to learn automatically from data as well as recognize patterns in the data that are subsequently used in making informed, rational decisions without explicit programming. Continuous improvement in performance is owed to experience and large dataset exposure; thus, ML models are an indispensable component in the advancement of AI technologies (5). The enhanced capacity to process data and improved computational ability has enhanced the capacity of different systems to formulate different artificially intelligent algorithms that can be incorporated in the clinical care pathway for conditions like AF (6). Different ML algorithms have been formulated that can substantially improve the detection and management of AF therefore helping to optimize the care pathway for AF (7).

Several studies have attempted to use AI for the early detection of AF using data from ECG, wearable technologies such as smartwatches and chest radiographs. The aim of this review is to summarize the findings of the studies attempted to use AI for the detection of AF. Through this narrative review, we intend to identify relevant and synthesize insights from the published literature to understand the role of AI in the management of AF. The findings from the evidence synthesized from this narrative review will help clinicians, data scientists and other relevant decision-makers to inculcate relevant findings into the clinical decision support systems for managing AF using AI. We present this article in accordance with the Narrative Review reporting checklist (available at https://jmai.amegroups.com/article/view/10.21037/jmai-24-278/rc).

Methods

For this narrative review, a comprehensive search was carried out to identify relevant articles from literature databases that included Google Scholar, PubMed and Medline. Different keywords that included “Atrial Fibrillation”, “AF”, “Artificial Intelligence”, “AI”, “Machine Learning”, and “Deep learning” with Boolean operators “OR” and “AND” were used to identify relevant articles for this review article. We included published studies of any study design that shared insights pertinent to the use of AI or ML for management of AF (Table 1). We excluded studies that were not in English or that were irrelevant to the topic. All the identified articles were reviewed by a single reviewer. After a detailed review of the titles and abstract of the identified articles, full articles were included in the study after assessing the eligibility criteria. Data from the identified articles was curated in a tabular format to show the details about the article and theme of the identified article. A total of 1,901 articles were identified out of which 185 articles were identified based on the title, 46 articles were selected based on the abstract of the studies and finally, 29 studies were included in the narrative review after reading the full-length articles (Figure 1).

Figure 1 PRISMA flow diagram for the study.

Results

The thorough search and evaluation based on our inclusion and exclusion criteria resulted in 29 studies that matched all criteria. It was observed that AI models have been tested using various datasets including ECG, Wearables and images of chest radiographs. The highest number of studies tested the ECG data. The number of studies that used data from wearables was 5 while only two studies used radiograph data for the detection of AF. The ML models can be used to effectively detect or predict individuals with AF. Understanding these models can play a pivotal role in formulating effective ML algorithms integrated into a clinical decision support system which can, therefore, be instrumental in reducing morbidities or mortality related to AF.

Detection of AF using AI

Recently many studies have reported the use of AI and ML in the detection and prediction of AF. Majorly patient data including age, gender and medical conditions with the clinical parameters from the electronic health records have been found to have predictive ability of AF. The studies use various approaches for the detection of AF by either using data of ECG, or from wearables. Some also used radiograph data to validate their models. The results of studies using different data sets are mentioned below (Table 2).

Use of ECG for AF detection

ECG has been used by clinicians for the last century to diagnose arrhythmia and other cardiac abnormalities. Although it is a very useful technology, the differential diagnosis is reliant on the reading ability of different physicians. At times, it can be very difficult to consider small variations and can lead to misdiagnosis. In the last couple of years, advancements in ML and AI made ECG reading automated and made it even more useful by enabling it to predict other cardiac dysfunctions and non-cardiac disorders such as cirrhosis. AI has the ability to observe the pattern change in large datasets with a lot of precision and accuracy. This makes ECG an ideal candidate for AI-based detection methods to read it precisely and accurately minimizing the human error of misdiagnosis.

Several studies reported the use of AI and ML-based tools to read ECG. One study tested the use of AI algorithms to improve the screening of previously unrecognized AF. This non-randomized interventional study conducted in 40 states of United States of America (USA), found that as compared to routine care, the use of an AI algorithm improved the detection of AF, therefore allowing an early detection and management of AF. Screening for AF through prolonged monitoring is costly, therefore, there is a need to formulate a point-of-care test to identify individuals with AF. AI-guided screening detected more AF than usual care, identifying it in 10.6% of high-risk and 2.4% of low-risk patients (8).

Another research group used convolutional neural network to formulate AI-enabled ECG to detect AF during sinus rhythms. This AI-enabled ECG was quite useful as it correctly screened individuals with AF with a sensitivity of 79% and specificity of 79.5% (9). Another study demonstrated the ability of deep learning algorithms to detect new onset of AF, demonstrating the ability of these algorithms to facilitate early identification of AF and therefore prevent stroke in these individuals. The model was able to predict new-onset AF within 1 year, with an AUC of 85% and an AUPRC of 22%. The model showed 69% sensitivity and 81% specificity during simulations. The model was able to identify 62% of high-risk patients who had an AF-related stroke within 3 years of the initial ECG (10).

One common problem in such algorithms is their low positive predictive value (PPV). One study reported that PPV of an implantable loop recorder was improved from 53.9% to 74.5% by using 2-part AI filter which used a deep neural network (11). Some studies have also found that AI-ECG analysis of AF can also be used to predict cerebral infarcts and cognitive decline. Based on a cohort study including 3,729 patients with the median age of 74 years, it was found that there was a significant correlation between AF score and infarcts therefore this research suggested a common precursor for AF and cognitive decline (12). A study was conducted to compare the predictive ability of AI-ECG and CHARGE-AF scores (14). It was found that when AI-ECG was combined with CHARGE-AF the discriminatory ability was improved to 72%. Additionally, it was advocated by the researchers that AI-ECG may be better than CHARGE-AF to predict AF because there is no requirement to manually extract the data for AI-ECG. Additionally, it is also possible to formulate better AF prediction scores as compared to CHARGE-AF score recommended in the current evidence by using ML on large contextually relevant datasets (14). This study recommended that the use of AI-ECG facilitates the clinical workflow because of ease of integration and automation as opposed to CHARGE-AF which requires extensive data abstraction (14). Such scores will enable the use of these models more effectively in primary healthcare settings and therefore scaling up the impact of these algorithms. One study found that AI-ECG can be used to predict AF among individuals having Graves’ disease with reduced ejection fraction. This finding is supported by a study that compared the predictive ability of traditional scores that included CHADS₂ (congestive heart failure, hypertension, age 75 years or older, diabetes mellitus, and stroke/transient ischemic attack) and CHA₂DS₂-VASc (congestive heart failure, hypertension, age 75 years or older, diabetes mellitus, stroke/transient ischemic attack, vascular disease, age 65 to 74 years, and sex category female) combined with AF burden using a model based on deep learning (15). This study reported a better predictive ability by deep neural network model as compared to the traditional risk scores to predict post-stroke AF. These ML algorithms have also the ability to predict the recurrence of AF after treatment with catheter ablation (16). Convolutional neural networks were reported to be optimal for identifying individuals with adverse possible outcomes after catheter ablation. A similar prediction of risk of AF recurrence can also be done for patients who undergo thoracoscopic surgical ablation. Some ML models can be used to predict non-pulmonary vein triggers before ablation (17,18).

Another study showed that deep learning approaches were better than traditional algorithms, enabling the clinicians to efficiently diagnose the patients with AF and therefore, reducing cost of care (19). Another study focusing on veterans’ database also demonstrated that deep learning-based models on ECG data had the ability to predict AF and thus supported the use of these models for screening for AF. These deep learning models have the ability to detect hidden patterns within the apparently normal sinus rhythm which have high probability of getting missed otherwise (20).

AF detection using wearable technology

Recent advancements in technology introduced many wearable consumer devices that are used to monitor health conditions such as heart rate and blood pressure. Many such devices are available in the market for detecting AF (21). However, wrist-worn LED sensors are less accurate due to heart rate underestimation. ECG-sensing devices, though less comfortable, provide verifiable ECG traces and enable more reliable AF detection by cardiologists and automated algorithms (22). Studies conducted to check the accuracy of such wearable devices resulted in variable results. One such study was conducted to check if low-cost consumer chest heart rate monitor devices can be used to accurately detect AF (Table 3). The support vector machine algorithm demonstrated excellent performance in detecting AF using RR interval data. In the training set, it achieved 99.2% sensitivity and 99.5% specificity, accurately identifying most AF cases and correctly classifying non-AF cases. In the validation set, it achieved 100% sensitivity by correctly identifying all 79 AF cases, and 97.6% specificity by correctly classifying 328 out of 336 non-AF cases. These results highlight the algorithm’s high accuracy and potential effectiveness in detecting AF (23). Such a device can, therefore, be used to effectively monitor and subsequently detect individuals with AF. There is a need to check the performance of this algorithm in a real-world population which will give insights about the cost-effectiveness of this algorithm and subsequently the risk stratification of the individuals for earlier interventions.

Smart wristband devices equipped with photoplethysmographic, and single-channel ECG systems can be used to detect AF. One study done to check the accuracy of such a device to detect AF found that the ML model built based on a photoplethysmographic device had sensitivity and specificity of 88.00% and 96.41% respectively. In the same way that ML models based on wristband ECG readings to detect AF showed an ability to detect AF with sensitivity of 87.33% and specificity of 99.20% (25). PPG readings had a sensitivity of 88.00%, specificity of 96.41%, and accuracy of 93.27%, while ECG readings showed 87.33% sensitivity, 99.20% specificity, and 94.76% accuracy. Physician-evaluated ECG records were even more accurate, with 96.67% sensitivity, 98.01% specificity, and 97.51% accuracy. Although the study showed promising results, the researchers only validated a single device on a small set of patients during inactivity.

One problem that was identified as a result of these wearable devices was that upon review by the physicians, the findings based on the prediction of these wearable devices yielded inconclusive results (24). To tackle this problem one study was conducted to compare the inconclusive findings between different types of wearable devices used to detect AF with deep neural network-based algorithm used in these devices. This study concluded that the use of deep neural network-based algorithm reduced the inconclusive findings while keeping the accuracy at the same level (27).

Another study which was done to use a deep learning model and leverage the data coming from photoplethysmographic signals coming from wearable devices to detect AF found that an accuracy of 99% for this model to detect AF. Due to this high accuracy, there is a need to check the feasibility of using these wearable devices in clinical settings. A similar finding came from a study that used deep learning to predict AF using photoplethysmographic data and found high sensitivity, specificity, PPV, and negative predictive value therefore outperforming other methods to predict AF (26).

Use of chest radiograph data for AF detection

Chest radiography is a common and simple procedure used for monitoring pulmonary diseases. Apart from their role in the diagnosis of pulmonary conditions chest radiographs can also be used to detect cardiac diseases. As mentioned earlier for AF diagnosis ECG is commonly used however there are chances that AF goes undiagnosed using ECG in around 13% of the cases (28). In some cases, a chest radiograph is a better choice for the detection of AF leading to efficient detection and early care for patients. Other than ECG and wearable data, radiographs have also been examined to check their potential for the detection of AF using AI (Table 4). A study reported that chest radiographs can also be used to identify individuals with AF. A deep learning model was formulated that could identify individuals with AF with an area under the curve of 81.26%. These chest radiographs were identified within 30 days of echocardiography. Based on the findings from medical records, these records were identified as AF positive and AF negative, and subsequently a deep learning model was trained to classify the radiographs into positive or negative for AF. The sensitivity for this model was 76% while specificity in this condition was 75% therefore, there are concerns related to false negatives identified in this process. However, the tested AI models in the study lack full transparency, as they cannot pinpoint whether they focus on the left atrium or nearby structures like the descending aorta or cardiac border. The AI model is more sensitive to permanent AF cases compared to paroxysmal AF, likely due to anatomical differences (29). In another study, the researchers tried to compare the accuracy of AI models to predict AF using chest radiographs with the residents of radiology to evaluate the accuracy of AI models. They analyzed 1,998 radiographs and found no difference in the detection between AI and radiology residents. However, AI showed higher specificity and PPV. The study suggests that a well-trained AI algorithm can match radiology residents in identifying chest radiograph findings (30). Detecting AF from chest radiographs is interesting academically but clinical utility is questionable. One possibility may be the use of such tools to support the screening of individuals for AF, requiring further confirmatory diagnostic evaluation.

Optimization of AF management

One group of researchers advocated that there is a need to optimize the use of anticoagulants among patients having AF using different ML models (31). Once AF is diagnosed in a patient, an important part of subsequent management is the decision to start anti-coagulation therapy to reduce the risk of stroke. This is a complicated decision as anti-coagulation therapies come with their own set of risk factors, especially, the increased risk of major bleeding, therefore further monitoring or testing may be required to perform a thorough risk vs. benefit analysis of such therapies. Since patients with AF may require lifelong treatment with oral anticoagulants (OACs), there is a need to identify the right choice of OACs for different situations. One analysis done on around 34,000 patients used a causal ML model to identify different patient subgroups with different outcomes associated with OAC use. This analysis underscores the importance of combining the use of data from AF patients with the use of ML models to optimize the choice of drug use and the dose of that drug (Table 5) (31).

Patients with AF are at an increased risk of stroke. Traditionally different algorithms like CHA₂DS₂-VASc score have been used to find out the risk of these individuals getting stroke. One retrospective study conducted on veteran data reported that the use of ML models like ensemble has a better ability to risk stratify AF patients in terms of their risk of getting stroke (32).

There is a substantial diversity among AF patients in the context of the factors that are required to make a decision related to the administration of OACs to these individuals. Considering that there are different types of OACs, and therefore there is a need to consider the heterogeneity of the patients to decide about the choice of anticoagulants for the individual patients. One study reported that causal ML models are useful in identifying patient subgroups and therefore, recommending the type and dose of anticoagulants for different groups (31).

Use of ML in primary care

Mostly there is a limitation of resources in the primary care centers and therefore, the ML algorithm formulated based on based on historical records can be used to risk stratify and therefore, predict AF patients in these resource-constrained centers. One such used data from around 2.9 million individuals to formulate ML to predict AF by incorporating factors recorded at the baseline as well time-varying factors (15). The ultimate utility of this and other similar algorithms can be tested after conducting randomized controlled trials to check whether the use of these algorithms improves patient outcomes or not in the real world.

Recurrence of AF after treatment

After treatment of AF patients with ablation therapy there is a possibility of recurrence of AF. ML models are formulated to predict this possible recurrence among AF patients after ablation (33). Several limitations were encountered during the analysis of this data. First of all, a small sample size was used in this analysis. The intensity profile of the CT images that were used in this study was not consistent which may have affected the findings from this study. Additionally, it was reported that the findings may be due to random variations like variations in ablation techniques and timing of the rhythm recording. It is therefore recommended to conduct such studies in a more controlled manner including a much larger number of study participants.

Another study was done to check the late recurrence of AF (33). This study reported that XGboost was the best model with an area under the curve of 75%. Important predictors that were identified in this context included thyroid stimulating hormone, recurrence of AF, and several blood markers that were measured before and after ablation. An important implication of this finding is that clinicians will tailor the management plan and therefore improve patient outcomes.

Prediction of outcomes of electrical cardioversion

One retrospective cohort study was done to assess the feasibility genderspecific prediction model to predict the success of electrical cardioversion for restoring sinus rhythm in among patients with AF. This study reported the use of logistic regression and other ML models to predict the success of cardioversion among patients of AF. Both logistic regression and ML model showed modest predictive ability to predict the restoration of sinus. The study highlighted the importance of personalized treatment approaches and the importance of taking patient specific factors into consideration while clinical decision-making (34). Through one study it was found that individuals who were enrolled in randomized controlled trials and had heart failure with reduced ejection fraction, in these subgroups of patients were identified as having AF who could potentially benefit from beta blockers. Therefore, this study identified potential methods to classify cluster patients into groups that have a better probability of benefiting from beta blockers (35).

Conclusions

AI has the potential to bring a revolution to medical sciences. The data revealed accurate detection of AF using various ML algorithms. ECG was found to be one of the most important modalities that can be used for AF early detection using AI. AI has demonstrated the ability to predict AF from sinus rhythm on an ECG and potentially detect AF from chest X-rays, the clinical applications of these remain limited, however, these will likely find a place in the screening process and in supporting the diagnosis. AI models can also assist in tailoring the management of patients with AF by predicting recurrence after cardioversion and in deciding on medical therapy. However, further studies are required for the validation of such models. Despite many limitations of the conducted studies, AI has a strong potential to revolutionize medicine and bring benefits to patients in terms of diagnosis and management of AF.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://jmai.amegroups.com/article/view/10.21037/jmai-24-278/rc

Peer Review File: Available at https://jmai.amegroups.com/article/view/10.21037/jmai-24-278/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jmai.amegroups.com/article/view/10.21037/jmai-24-278/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Lippi G, Sanchis-Gomar F, Cervellin G. Global epidemiology of atrial fibrillation: An increasing epidemic and public health challenge. Int J Stroke 2021;16:217-21. [Crossref] [PubMed]
2. Lip GYH, Brechin CM, Lane DA. The global burden of atrial fibrillation and stroke: a systematic review of the epidemiology of atrial fibrillation in regions outside North America and Europe. Chest 2012;142:1489-98. [Crossref] [PubMed]
3. Alonso A, Krijthe BP, Aspelund T, et al. Simple risk model predicts incidence of atrial fibrillation in a racially and geographically diverse population: the CHARGE-AF consortium. J Am Heart Assoc 2013;2:e000102. [Crossref] [PubMed]
4. Prasitlumkum N, Cheungpasitporn W, Chokesuwattanaskul A, et al. Diagnostic accuracy of smart gadgets/wearable devices in detecting atrial fibrillation: A systematic review and meta-analysis. Arch Cardiovasc Dis 2021;114:4-16. [Crossref] [PubMed]
5. Kufel J, Bargieł-Łączek K, Kocot S, et al. What Is Machine Learning, Artificial Neural Networks and Deep Learning?-Examples of Practical Applications in Medicine. Diagnostics (Basel) 2023;13:2582. [Crossref] [PubMed]
6. Ono M, Varma N. Remote Monitoring for Chronic Disease Management: Atrial Fibrillation and Heart Failure. Card Electrophysiol Clin 2018;10:43-58. [Crossref] [PubMed]
7. Hamet P, Tremblay J. Artificial intelligence in medicine. Metabolism 2017;69S:S36-40. [Crossref] [PubMed]
8. Noseworthy PA, Attia ZI, Behnken EM, et al. Artificial intelligence-guided screening for atrial fibrillation using electrocardiogram during sinus rhythm: a prospective non-randomised interventional trial. Lancet 2022;400:1206-12. [Crossref] [PubMed]
9. Attia ZI, Noseworthy PA, Lopez-Jimenez F, et al. An artificial intelligence-enabled ECG algorithm for the identification of patients with atrial fibrillation during sinus rhythm: a retrospective analysis of outcome prediction. Lancet 2019;394:861-7. [Crossref] [PubMed]
10. Raghunath S, Pfeifer JM, Ulloa-Cerna AE, et al. Deep Neural Networks Can Predict New-Onset Atrial Fibrillation From the 12-Lead ECG and Help Identify Those at Risk of Atrial Fibrillation-Related Stroke. Circulation 2021;143:1287-98. [Crossref] [PubMed]
11. Mittal S, Oliveros S, Li J, et al. AI Filter Improves Positive Predictive Value of Atrial Fibrillation Detection by an Implantable Loop Recorder. JACC Clin Electrophysiol 2021;7:965-75. [Crossref] [PubMed]
12. Weil EL, Noseworthy PA, Lopez CL, et al. Artificial Intelligence-Enabled Electrocardiogram for Atrial Fibrillation Identifies Cognitive Decline Risk and Cerebral Infarcts. Mayo Clin Proc 2022;97:871-80. [Crossref] [PubMed]
13. Melzi P, Tolosana R, Cecconi A, et al. Analyzing artificial intelligence systems for the prediction of atrial fibrillation from sinus-rhythm ECGs including demographics and feature visualization. Sci Rep 2021;11:22786. [Crossref] [PubMed]
14. Christopoulos G, Graff-Radford J, Lopez CL, et al. Artificial Intelligence-Electrocardiography to Predict Incident Atrial Fibrillation: A Population-Based Study. Circ Arrhythm Electrophysiol 2020;13:e009355. [Crossref] [PubMed]
15. Hill NR, Ayoubkhani D, McEwan P, et al. Predicting atrial fibrillation in primary care using machine learning. PLoS One 2019;14:e0224582. [Crossref] [PubMed]
16. Naser JA, Attia ZI, Pislaru SV, et al. Artificial Intelligence Application in Graves Disease: Atrial Fibrillation, Heart Failure and Menstrual Changes. Mayo Clin Proc 2022;97:730-7. [Crossref] [PubMed]
17. Zheng X, Wang F, Zhang J, et al. Using machine learning to predict atrial fibrillation diagnosed after ischemic stroke. Int J Cardiol 2022;347:21-7. [Crossref] [PubMed]
18. Zhou X, Nakamura K, Sahara N, et al. Deep Learning-Based Recurrence Prediction of Atrial Fibrillation After Catheter Ablation. Circ J 2022;86:299-308. [Crossref] [PubMed]
19. Hwang YT, Lee HL, Lu CH, et al. A Novel Approach for Predicting Atrial Fibrillation Recurrence After Ablation Using Deep Convolutional Neural Networks by Assessing Left Atrial Curved M-Mode Speckle-Tracking Images. Front Cardiovasc Med 2020;7:605642. [Crossref] [PubMed]
20. Baalman SWE, Lopes RR, Ramos LA, et al. Prediction of Atrial Fibrillation Recurrence after Thoracoscopic Surgical Ablation Using Machine Learning Techniques. Diagnostics (Basel) 2021;11:1787. [Crossref] [PubMed]
21. Perez MV, Mahaffey KW, Hedlin H, et al. Large-Scale Assessment of a Smartwatch to Identify Atrial Fibrillation. N Engl J Med 2019;381:1909-17. [Crossref] [PubMed]
22. Koshy AN, Sajeev JK, Nerlekar N, et al. Smart watches for heart rate assessment in atrial arrhythmias. Int J Cardiol 2018;266:124-7. [Crossref] [PubMed]
23. Lown M, Brown M, Brown C, et al. Machine learning detection of Atrial Fibrillation using wearable technology. PLoS One 2020;15:e0227401. [Crossref] [PubMed]
24. Chen E, Jiang J, Su R, et al. A new smart wristband equipped with an artificial intelligence algorithm to detect atrial fibrillation. Heart Rhythm 2020;17:847-53. [Crossref] [PubMed]
25. Kwon S, Hong J, Choi EK, et al. Deep Learning Approaches to Detect Atrial Fibrillation Using Photoplethysmographic Signals: Algorithms Development Study. JMIR Mhealth Uhealth 2019;7:e12770. [Crossref] [PubMed]
26. Poh MZ, Poh YC, Chan PH, et al. Diagnostic assessment of a deep learning system for detecting atrial fibrillation in pulse waveforms. Heart 2018;104:1921-8. [Crossref] [PubMed]
27. Mannhart D, Lefebvre B, Gardella C, et al. Clinical validation of an artificial intelligence algorithm offering cross-platform detection of atrial fibrillation using smart device electrocardiograms. Arch Cardiovasc Dis 2023;116:249-57. [Crossref] [PubMed]
28. Benjamin EJ, Muntner P, Alonso A, et al. Heart Disease and Stroke Statistics-2019 Update: A Report From the American Heart Association. Circulation 2019;139:e56-e528. Erratum in: Circulation 2020;141:e33. [Crossref] [PubMed]
29. Matsumoto T, Ehara S, Walston SL, et al. Artificial intelligence-based detection of atrial fibrillation from chest radiographs. Eur Radiol 2022;32:5890-7. [Crossref] [PubMed]
30. Wu JT, Wong KCL, Gur Y, et al. Comparison of Chest Radiograph Interpretations by Artificial Intelligence Algorithm vs Radiology Residents. JAMA Netw Open 2020;3:e2022779. [Crossref] [PubMed]
31. Ngufor C, Yao X, Inselman JW, et al. Identifying treatment heterogeneity in atrial fibrillation using a novel causal machine learning method. Am Heart J 2023;260:124-40. [Crossref] [PubMed]
32. Han L, Askari M, Altman RB, et al. Atrial Fibrillation Burden Signature and Near-Term Prediction of Stroke: A Machine Learning Analysis. Circ Cardiovasc Qual Outcomes 2019;12:e005595. [Crossref] [PubMed]
33. Atta-Fosu T, LaBarbera M, Ghose S, et al. A new machine learning approach for predicting likelihood of recurrence following ablation for atrial fibrillation from CT. BMC Med Imaging 2021;21:45. [Crossref] [PubMed]
34. Vinter N, Frederiksen AS, Albertsen AE, et al. Role for machine learning in sex-specific prediction of successful electrical cardioversion in atrial fibrillation? Open Heart 2020;7:e001297. [Crossref] [PubMed]
35. Karwath A, Bunting KV, Gill SK, et al. Redefining β-blocker response in heart failure patients with sinus rhythm and atrial fibrillation: a machine learning cluster analysis. Lancet 2021;398:1427-35. [Crossref] [PubMed]