Computer Science Meets Squiggly Lines: How Artificial Intelligence Will Revolutionize Atrial Fibrillation Care
© 2024 HMP Global. All Rights Reserved.
Any views and opinions expressed are those of the author(s) and/or participants and do not necessarily reflect the views, policy, or position of EP Lab Digest or HMP Global, their employees, and affiliates.
EP LAB DIGEST. 2024;24(5):1,8-9.
Artificial intelligence (AI) is attracting tremendous amounts of attention in medicine today. As computer science has enjoyed groundbreaking advances that have transformed other sectors of society, the idea of its application in medicine has generated excitement over its potential to fill knowledge gaps, paving the way towards precision medicine. Cardiac electrophysiology (EP) is one of the most well-situated specialties to reap tremendous gains in AI. With various publications touting its remarkable ability to predict heart failure risk, ejection fraction, sudden death, and response to cardiac resynchronization therapy from a 12-lead electrocardiogram (ECG) alone, it is expected this will be a dominant vehicle of resolving unknowns and revolutionizing our field.1-3 One source of unanswered questions is in the area of atrial fibrillation (AF).
Amid the myriad pathologies electrophysiologists contend with daily, AF rises above all others by sheer numbers and unrelenting socioeconomic burden of disease. Battling AF is exigent yet complex, riddled with unknowns in proper detection, meaningful prognostication, evidence-based management, and effective prevention of its devastating consequences.4 Yet, as the burden of disease and data deluge increases, hidden within this is an opportunity. Understanding the basics of AI can help identify areas where this innovative tool can transform the future of AF care.
What Is AI?
While the term was coined by Stanford professor John McCarthy in 1955, the concept was first introduced 5 years earlier by English mathematician and computer scientist Alan Turing.5,6 Turing pioneered the novel idea that machines could simulate human intelligence. At its core, AI is the combination of statistical methods, computational science, and human neurophysiology to automate human tasks. Such tasks, like pattern recognition, text summarization/synthesis, and decision-making, are thus mirrored by their computational analogues: machine learning (ML), natural language processing (NLP), computer vision, and robotics.1,6
Major breakthroughs in ML, the learning and decision-making domain of AI, have enhanced its potential to transform cardiovascular medicine. In ML, an algorithm is designed to “learn” from input data to predict outputs or observe patterns within the input data itself. These concepts are not new. In fact, the regression analysis methods generating the Framingham and CHA2DS2-VASc risk scores are considered simple examples of classical ML.7 One ML method called supervised learning requires a human to preselect input features and output labels upon which the algorithm is trained, with the goal of predicting a known output as well as a human.8,9 Alternatively, unsupervised learning allows an algorithm to independently recognize important features, by observing patterns that even humans cannot observe. A subtype of ML is deep learning (DL), which integrates supervised and unsupervised learning and utilizes artificial neural networks (ANNs). Inspired by the architecture of human neurons and synapses, ANNs consist of interconnected multilayered networks of statistically-weighted nodes, or artificial neurons, through which data is processed, providing highly accurate labels and feature extraction, which are then validated by a supervised approach.7,8 These methods have accelerated the field of computer science and transformed other AI disciplines, such as NLP and computer vision, as the need for humans to know and preassign features, labels, and rules is taken out of the equation. An example is the ability to determine ejection fraction from a 12-lead ECG.3 It is no surprise that ML is of great interest in health care, as it is grounded in the tasks of critical thinking and decision-making that are crucial in medicine.
AI and the Electrophysiologist
What allows electrophysiologists to be uniquely positioned to leverage AI is that EP tools lend themselves well for training an effective model. An excellent example is the ECG. Electrocardiographic tracings are easily obtainable in many formats including standard 12-lead ECGs, continuous Holter monitoring, photoplethysmography (PPG), and single-lead ECG wearables.10 ANNs can detect subtle changes in raw ECG signals that humans are unable to observe. Furthermore, device and intraprocedural electrograms (EGMs), cardiac imaging, and even text in electronic charts, offer a wealth of data that is leverageable to automate tasks and solve dilemmas. However, accessibility to large quality datasets representative of the general population is required for meaningful outputs.7,10
Challenges and Opportunities in AF
There are several unknowns in AF, including adequate screening, identifying patients at risk for AF-associated morbidity, and discerning ideal therapeutic strategies, that ML can help address (Figure 1).
Both detection and prediction of AF is challenging. Many patients are asymptomatic, leaving them undiagnosed until their first AF-associated complication (eg, stroke or heart failure). ML-enabled direct-to-consumer PPGs and ECG wearables have made early AF diagnosis incredibly accessible.11-14 Moreover, predicting short-term incident AF in those without an established diagnosis is possible by utilizing a DL model with both 12-lead and ambulatory single-lead patch monitor ECGs in sinus rhythm.15,17 Even NLP-enabled electronic health records (EHRs) can sort through large bodies of text and accurately identify individuals at risk for developing AF better than traditional risk calculators.18
What about preventing stroke? Its relationship with AF is still unclear, including the temporality and characterization of AF that precedes it and patient phenotypes most at risk. The CHA2DS2-VASc score is imperfect, yet it is the foundation of major therapeutic decisions.7 The ever-increasing diagnoses of AF via wearables adds a layer of complexity. How should one approach a single short smartwatch recording of AF, or short high-rate atrial events on device interrogations? Does a patient with postoperative AF need lifelong oral anticoagulant (OAC) therapy? Who will have a stroke in the future? Electrophysiologists are just beginning to scratch the surface in answering these questions. A follow-up to the Huawei Heart Study suggested early AF diagnosis as key in decreasing stroke or thromboembolism, death, and hospitalization with PPGs.19 Smaller studies are characterizing device-detected AF burden and associated 30-day stroke, while others demonstrated stroke prediction by 12-lead ECG signals and epicardial adipose tissue volume on cardiac CT.20-22 In the future, risk stratification methods will be refined, allowing for the construction of sophisticated predictive models to inform a more personalized OAC strategy. Feasibility and efficacy of a pill-in-the-pocket approach or traditional daily dosing will be ascertained. Compounding imaging, ECG, and clinical data in highly-trained models may better inform decision-making in choosing OAC vs left atrial appendage occlusion.
Identifying those who would most benefit from ablation and/or antiarrhythmic drugs (AADs) is also an area that can be investigated with DL in particular. Clinical trials such as CASTLE-AF and EAST-AFNET 4 suggest the benefits of upfront rhythm control, and trials such as CABANA suggest that catheter ablation may be superior to AAD therapy.4 However, such data are difficult to generalize given the heterogeneity in presentation and treatment response. While there have been tremendous efforts towards understanding mechanisms and studying ablation strategies for AF, mechanistic knowledge beyond the pulmonary veins is still incomplete. Recurrence rates remain between 20%-45%.23 Yet there is fertile ground for precision medicine to bloom. Opportunity exists in leveraging the use of digital twins (DTs) to inform personalized management strategies. Intraprocedural EGMs employed in DL algorithms can demystify AF mechanisms. Early studies illustrated the accuracy of convolutional neural networks (CNNs) to locate drivers and rotors using EGMs and CMR data, while others have used DTs to generate personalized ablation strategies, targeting regions of high dominant frequencies.24-27 This, of course, would have to be replicated in vivo in future studies. ML methods may predict recurrences after ablation and cardioversion. Moreover, the optimal choice of AAD when appropriate, as well as drug dosing and adjustment, may be informed by the integration of DTs and ML. Wearable data in conjunction with cloud-based AI-enabled QTc analysis can allow safe, effective AAD loading even in the comfort of one’s home.
The Digital Health Clinic
Uncertainties in clinical management and mounting data deluge of EHRs, remote transmissions, and wearables is daunting. As wearables in the general population become more ubiquitous, patients desire more interaction with clinicians, seeking guidance on interpreting data supported by little evidence. Here is where the digital health clinic may play a pivotal role (Figure 2).10 With novel infrastructure and the right allocation of resources, patients can feel free to interact and clinic staff will be empowered to engage effectively. Streamlined workflows and precision medicine may be possible with the structuring of an AI-enabled patient platform integrated into an EHR. Ambulatory monitoring alerts and reports can be filtered and annotated before being sent to clinical staff for review. Clerical work that takes away from patient care can be automated. Personalized risk assessment and care plans can be generated with patient clinical data. A well-designed clinic can address the influx of new patients prompted by the use of wearables, as well as implement strong AF management programs with effective AAD and postprocedural monitoring. It will take time to surmount the barriers, but the future is bright.
The Future Is at Hand
AI is changing the fabric of our society. Medicine is rife with conundrums that innovative predictive modeling and analytics can solve. The management of the AF patient is an example of this. Challenges remain in the implementation of ML models, yet these novel technologies have tremendous potential to streamline AF diagnosis and treatment, blazing a path towards precision medicine and personalized care.
Acknowledgement
The author wishes to thank Dr Hamid Ghanbari, who has been nothing short of being a phenomenal mentor.
Disclosure: Dr Devgun has completed and returned the ICMJE Form for Disclosure of Potential Conflicts of Interest, and reports no conflicts of interest regarding the content herein. She discloses she is a Fellow-In-Training Editor of the Cardiovascular Digital Health Journal (unpaid).
References
1. Kabra R, Israni S, Vijay B, et al. Emerging role of artificial intelligence in cardiac electrophysiology. Cardiovasc Digit Health J. 2022;3(6):263-275. doi:10.1016/j.cvdhj.2022.09.001
2. Butler L, Karabayir I, Kitzman DW, et al. A generalizable electrocardiogram-based artificial intelligence model for 10-year heart failure risk prediction. Cardiovasc Digit Health J. 2023;4(6):183-190. doi:10.1016/j.cvdhj.2023.11.003
3. Attia ZI, Kapa S, Lopez-Jimenez F, et al. Screening for cardiac contractile dysfunction using an artificial intelligence-enabled electrocardiogram. Nat Med. 2019;25(1):70-74. doi:10.1038/s41591-018-0240-2
4. Joglar JA, Chung MK, Armbruster AL, et al. 2023 ACC/AHA/ACCP/HRS Guideline for the Diagnosis and Management of Atrial Fibrillation: a report of the American College of Cardiology/American Heart Association Joint Committee on Clinical Practice Guidelines. Circulation. 2024;149(1):e1-e156. doi:10.1161/CIR.0000000000001193
5. Mintz Y, Brodie R. Introduction to artificial intelligence in medicine. Minim Invasive Ther Allied Technol. 2019;28(2):73-81. doi:10.1080/13645706.2019.1575882
6. Hamet P, Tremblay J. Artificial intelligence in medicine. Metabolism. 2017;69S:S36-S40. doi:10.1016/j.metabol.2017.01.011
7. Siontis KC, Yao X, Pirruccello JP, Philippakis AA, Noseworthy PA. How will machine learning inform the clinical care of atrial fibrillation? Circ Res. 2020;127(1):155-169. doi:10.1161/CIRCRESAHA.120.316401
8. Deo RC. Machine learning in medicine. Circulation. 2015;132(20):1920-1930. doi:10.1161/CIRCULATIONAHA.115.001593
9. Beam AL, Kohane IS. Big data and machine learning in health care. JAMA. 2018;319(13):1317-1318. doi:10.1001/jama.2017.18391
10. Varma N, Han JK, Passman R, et al. Promises and perils of consumer mobile technologies in cardiovascular care: JACC Scientific Statement. J Am Coll Cardiol. 2024;83(5):611-631. doi:10.1016/j.jacc.2023.11.024
11. Perez MV, Mahaffey KW, Hedlin H, et al. Large-scale assessment of a smartwatch to identify atrial fibrillation. N Engl J Med. 2019;381(20):1909-1917. doi:10.1056/NEJMoa1901183
12. Bumgarner JM, Lambert CT, Hussein AA, et al. Smartwatch algorithm for automated detection of atrial fibrillation. J Am Coll Cardiol. 2018;71(21):2381-2388. doi:10.1016/j.jacc.2018.03.003
13. Lubitz SA, Faranesh AZ, Selvaggi C, et al. Detection of atrial fibrillation in a large population using wearable devices: the Fitbit Heart Study. Circulation. 2022;146(19):1415-1424. doi:10.1161/CIRCULATIONAHA.122.060291
14. Poh MZ, Battisti AJ, Cheng LF, et al. Validation of a deep learning algorithm for continuous, real-time detection of atrial fibrillation using a wrist-worn device in an ambulatory environment. J Am Heart Assoc. 2023;12(19):e030543. doi:10.1161/JAHA.123.030543
15. 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(10201):861-867. doi:10.1016/S0140-6736(19)31721-0
16. Suzuki S, Motogi J, Umemoto T, et al. Lead-specific performance for atrial fibrillation detection in convolutional neural network models using sinus rhythm electrocardiography. Circ Rep. 2024;6(3):46-54. doi:10.1253/circrep.CR-23-0068
17. Gadaleta M, Harrington P, Barnhill E, et al. Prediction of atrial fibrillation from at-home single-lead ECG signals without arrhythmias. NPJ Digit Med. 2023;6(1):229. doi:10.1038/s41746-023-00966-w
18. Ashburner JM, Chang Y, Wang X, et al. Natural language processing to improve prediction of incident atrial fibrillation using electronic health records. J Am Heart Assoc. 2022;11(15):e026014. doi:10.1161/JAHA.122.026014
19. Guo Y, Guo J, Shi X, et al. Mobile health technology-supported atrial fibrillation screening and integrated care: a report from the mAFA-II trial long-term extension cohort. Eur J Intern Med. 2020;82:105-111. doi:10.1016/j.ejim.2020.09.024
20. 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(10):e005595. doi:10.1161/CIRCOUTCOMES.118.005595
21. 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(13):1287-1298. doi:10.1161/CIRCULATIONAHA.120.047829
22. West HW, Siddique M, Williams MC, et al. Deep-learning for epicardial adipose tissue assessment with computed tomography: implications for cardiovascular risk prediction. JACC Cardiovasc Imaging. 2023;16(6):800-816. doi:10.1016/j.jcmg.2022.11.018
23. Dretzke J, Chuchu N, Agarwal R, et al. Predicting recurrent atrial fibrillation after catheter ablation: a systematic review of prognostic models. Europace. 2020;22(5):748-760. doi:10.1093/europace/euaa041
24. Ravikumar V, Annoni E, Parthiban P, et al. Novel mapping techniques for rotor core detection using simulated intracardiac electrograms. J Cardiovasc Electrophysiol. 2021;32(5):1268-1280. doi:10.1111/jce.14948
25. Ríos-Muñoz GR, Fernández-Avilés F, Arenal Á. Convolutional neural networks for mechanistic driver detection in atrial fibrillation. Int J Mol Sci. 2022;23(8):4216. doi:10.3390/ijms23084216
26. Bai J, Lu Y, Wang H, Zhao J. How synergy between mechanistic and statistical models is impacting research in atrial fibrillation. Front Physiol. 2022;13:957604. doi:10.3389/fphys.2022.957604
27. Azzolin L, Eichenlaub M, Nagel C, et al. Personalized ablation vs. conventional ablation strategies to terminate atrial fibrillation and prevent recurrence. Europace. 2023;25(1):211-222. doi:10.1093/europace/euac116
