Skip to main content

Advertisement

ADVERTISEMENT

Cover Story

Artificial Intelligence in Clinical Cardiac Electrophysiology

Vidhushei Yogeswaran, MD1*; Jacob J. Mayfield, MD1*; Patrick M. Boyle, PhD2-4; Arun R. Sridhar, MBBS, MPH1

*VY and JJM are equal contributors to this publication. 1Department of Medicine, Division of Cardiology, University of Washington Medical Center, Seattle, Washington; 2Department of Bioengineering, University of Washington, Seattle, Washington; 3Institute for Stem Cell and Regenerative Medicine, University of Washington, Seattle, Washington; 4Center for Cardiovascular Biology, University of Washington, Seattle, Washington

February 2022
1535-2226

Artificial intelligence (AI) has permeated many aspects of everyday life. From online shopping to finance, and certainly in the medical literature, AI has become a buzzword for technological progress. AI refers to computer-based processing of information that would otherwise typically require human cognition.

Machine learning (ML) is a branch of AI that uses algorithms to identify patterns and relationships from datasets.1-3 Building on its success in other fields, AI approaches in cardiology have received significant attention over the last few years. AI techniques have already become a standard part of image processing in nuclear cardiology and there has been substantial growth in echocardiography AI research.1 Rapid growth of consumer devices producing medical data, coupled with recent advances in AI-augmented electrocardiography techniques, may herald new breakthroughs in the computational interpretation of ECGs.3 Rapid proliferation of available biological and clinical data as well as the push for personalized medicine have created a need for more effective and intuitive tools to empower cardiologists to make optimal management decisions for individual patients — ML has the potential to fill this niche.1,2,4

A relatively new and exciting type of ML, known as deep learning, involves deep neural networks (DNNs) that automate feature recognition.2 DNNs are artificial neural networks consisting of multiple processing layers that are interconnected and able to learn from their input data.3 DNNs can recognize subtle patterns and relationships without extensive preprocessing, and have demonstrated their utility in several challenges in cardiology. In electrophysiology, algorithms have been rising in popularity with studies showing their ability to improve arrhythmia detection, arrhythmia localization, ECG interpretation, and tailored drug dosing.1,5,6 Recently, the Apple Heart Study investigators showed that their algorithm had a high positive predictive value for the detection of atrial fibrillation with an irregular pulse detection.3,7 The DNN approach has also been used to predict potassium levels using ECG features in patients at high risk for dyskalemias.8,9

Teamwork Brings Results

With its mathematical foundations, advances in ML haves traditionally been driven by computer scientists.3 In order to leverage ML effectively in medicine, it is necessary to cultivate collaboration between clinicians, programming experts, and traditional statisticians, ensuring ML applications are clinically meaningful while remaining computationally and statistically feasible. Our group at the University of Washington (UW) is a partnership between experts from the College of Engineering and those in the School of Medicine, bringing together bioengineers, computer scientists, electrophysiologists, and trainees from various disciplines. Over the past few years, our team has created models to detect arrhythmias using smart speaker systems, worked towards sudden cardiac death (SCD) risk stratification and prediction, and trained and validated algorithms to detect the risk of complications in hospitalized patients with COVID-19. We are currently working on expanding our work, for example, combining deep learning augmented ECG and advanced imaging to predict clinical outcomes and continue studying contemporary challenges in cardiovascular disease.

Similar multidisciplinary collaborations exist at our peer institutions, including Mayo Clinic, Cleveland Clinic, Stanford University, and the University of California San Francisco. These groups have published work validating different algorithms to help further patient care in cardiology. Review articles by Feeney et al (2020), Trayanova et al (2021), and Lopez-Jimenez et al (2020) highlight several important studies.1-3

Several collaborative groups have evaluated how AI can be leveraged to address challenges posed by the COVID-19 pandemic. The highly transmissible novel coronavirus has infected millions of people and has led to more than 4 million deaths worldwide.10 Despite medical advancements, vaccinations, and public health efforts, uncontained outbreaks continue to overwhelm healthcare systems. COVID-19 is linked with a range of cardiovascular complications including acute coronary syndrome, myocarditis, thromboembolism, and life-threatening arrhythmia.11 In the context of healthcare systems functioning at contingency and even crisis levels of care, one of the most important challenges is rapid identification of patients at highest risk for adverse outcomes and triaging them to appropriate acuity-based care.

Over the last year, our team has developed a rapidly deployable DNN model that uses intake ECG alone to predict adverse outcomes in hospitalized patients with COVID-19 (Figure 2). We used data from 1386 patients hospitalized with COVID-19 at four different university hospitals — the University of Washington in Seattle, Karolinska Institutet University Hospital in Solna, Sweden, Uppsala University Hospital in Uppsala, Sweden, and Copenhagen University Hospital in Copenhagen, Denmark — to train two DNNs to predict major adverse cardiovascular events (MACE), defined as arrhythmic events, new onset heart failure, or thromboembolic events.12 Arrhythmic events included new-onset atrial fibrillation, sustained ventricular tachycardia, ventricular fibrillation, tachycardic and bradycardic arrests, and a high burden of premature ventricular complexes. Our DNNs yielded similar performance scores to other commonly used cardiovascular risk stratification tools, such as CHA₂DS₂-VASc, in predicting death and MACE in this population.13 This work was presented at Heart Rhythm 2021 and received an award for the highest scoring abstract in the Digital Health category.12,14

Sridhar Artificial Intelligence Figure 1Sridhar Artificial Intelligence Figure 2

Detecting Sudden Cardiac Death: One Algorithm at a Time

Prediction and early recognition of malignant arrhythmias is key to the prevention of SCD. ML has the potential to improve cardiac arrest survival, from predicting incident arrhythmias, to improving automated arrhythmia diagnosis and therapy delivery, to clinical decision-making support in the hospital. DNNs have demonstrated similar performance of arrhythmia ECG detection to that of cardiologists.5,6 Acute coronary syndromes may incite ventricular arrhythmias, leading to SCD.15 DNNs have the potential to improve early detection of acute coronary syndrome, utilizing technology already in routine use for in- and out-of-hospital cardiac arrest rescue efforts.1 Algorithms have shown promising results for predicting ventricular fibrillation and ventricular tachycardia.16,17 There are several centers working towards evaluating risk prediction tools for in-hospital cardiac arrests using various input parameters. Last year, a group developed and validated a DNN for predicting cardiac arrest using ECG alone.18 Additionally, deep learning-based algorithms developed using vital sign inputs have shown a high sensitivity and low false alarm rate for detecting in-hospital cardiac arrest.19

Recently, there have been exciting advances in the use of consumer devices to aid in early detection of SCD. Agonal breathing is common in patients suffering cardiac arrest.20 A multidisciplinary team at UW recently trained a DNN to identify agonal breathing patterns captured by voice recognition features built into consumer devices such as Apple iPhone, Amazon Alexa, and Samsung Galaxy products. The team built a contactless detection speaker system to effectively identify cardiac arrest-associated agonal breathing patterns. These technological advancements have the potential to significantly improve outcomes for victims of unwitnessed arrest.21

Where Are We Heading?

AI, and specifically ML, has shown great potential to improve many facets of cardiovascular medicine and electrophysiology. With the growing interest in the field and the increase in the biological and clinical data available, we anticipate that there will be a continued renaissance of ML-augmented tools to improve clinician effectiveness and personalize patient care. There remains ample opportunity to evaluate and validate different AI-algorithms as a risk prediction tool using the electronic health records, basic science, and clinical medicine for common high-risk conditions including influenza, pneumonia, and coronary artery disease. We hope that there will be future frameworks to incorporate AI into everyday clinical practice. 

Acknowledgments

This work was supported by a COVID-19 Rapid Response Grant from the University of Washington Population Health Initiative (awarded to ARS and PMB).

Disclosures: The authors have completed and returned the ICMJE Form for Disclosure of Potential Conflicts of Interest. The authors have no conflicts of interest to report regarding the content herein.

References

1. Lopez-Jimenez F, Attia Z, Arruda-Olson A, et al. Artificial intelligence in cardiology: present and future. Mayo Clin Proc. 2020;95(5):1015-1039.

2. Feeny AK, Chung MK, et al. Artificial intelligence and machine learning in arrhythmias and cardiac electrophysiology. Circ Arrhythm Electrophysiol. 2020;13(8):e007952.

3. Trayanova NA, Popescu DM, Shade JK. Machine learning in arrhythmia and electrophysiology. Circ Res. 2021;128(4):544-566.

4. Itchhaporia D, Snow PB, ALmassy RJ, Oetgen WJ. Artificial neural networks: current status in cardiovascular medicine. J Am Coll Cardiol. 1996;28(2):515-521.

5. Hannun AY, Rajpurkar P, Haghpanahi M, et al. Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nat Med. 2019;25(1):65-69.

6. van de Leur RR, Boonstra MJ, Bagheri A, et al. Big data and artificial intelligence: opportunities and threats in electrophysiology. Arrhythm Electrophysiol Rev. 2020;9(3):146-154.

7. 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.

8. Corsi C, Cortesi M, Callisesi G, et al. Noninvasive quantification of blood potassium concentration from ECG in hemodialysis patients. Sci Rep. 2017;7(1):42492.

9. Lin CS, Lin C, Fang WH, et al. A deep-learning algorithm (ECG12Net) for detecting hypokalemia and hyperkalemia by electrocardiography: algorithm development. JMIR Med Inform. 2020;8(3):e15931.

10. Hu B, Guo H, Zhou P, Shi ZL. Characteristics of SARS-CoV-2 and COVID-19. Nat Rev Microbiol. 2021;19(3):141-154.

11. Tajbakhsh A, Gheibi Hayat SM, Taghizadeh H, et al. COVID-19 and cardiac injury: clinical manifestations, biomarkers, mechanisms, diagnosis, treatment, and follow up. Expert Rev Anti Infect Ther. 2021;19(3):345-357.

12. Sridhar AR, Chen ZH, Mayfield JJ, et al. Identifying risk of adverse outcomes in COVID-19 patients via artificial intelligence-powered analysis of 12-lead intake electrocardiogram. Cardiovasc Digit Health J. 2021 Dec 31. Online ahead of print. doi: 10.1016/j.cvdhj.2021.12.003

13. Sanders GD, Lowenstern A, Borre E, et al. Stroke prevention in patients with atrial fibrillation: a systematic review update. Rockville (MD): Agency for Healthcare Research and Quality (US); 2018 Oct. Report No.: 18-EHC018-EFReport No.: 2018-SR-04.

14. Boyle P, Chen ZH, Zamponi AF, et al. B-PO01-094: Artificial intelligence (AI) can identify risk of death in COVID-19 patients using 12-lead intake electrocardiogram (ECG) alone. Heart Rhythm. 2021;18(8 Suppl):S88. doi: 10.1016/j.hrthm.2021.06.238

15. Adabag AS, Luepker RV, Roger VL, Gersh BJ. Sudden cardiac death: epidemiology and risk factors. Nat Rev Cardiol. 2010;7(4):216-225.

16. Taye GT, Shim EB, Hwang HJ, Lim KM. Machine learning approach to predict ventricular fibrillation based on QRS complex shape. Front Physiol. 2019;10:1193-1193.

17. Ebrahimzadeh E, Pooyan M, Bijar A. A novel approach to predict sudden cardiac death (SCD) using nonlinear and time-frequency analyses from HRV signals. PLoS One. 2014;9(2):e81896.

18. Kwon JM, Kim KH, Jeon KH, Lee SY, Park J, Oh BH. Artificial intelligence algorithm for predicting cardiac arrest using electrocardiography. Scand J Trauma Resusc Emerg Med. 2020;28(1):98.

19. Kwon JM, Lee Y, Lee Y, Lee S, Park J. An algorithm based on deep learning for predicting in-hospital cardiac arrest. J Am Heart Assoc. 2018;7(13):e008678.

20. Rea TD. Agonal respirations during cardiac arrest. Curr Opin Crit Care. 2005;11(3):188-191.

21. Chan J, Rea T, Gollakota S, Sunshine JE. Contactless cardiac arrest detection using smart devices. NPJ Digit Med. 2019;2(1):52.


Advertisement

Advertisement

Advertisement