Abstract
Background
Current electrocardiogram analysis algorithms cannot predict the presence of coronary
artery disease (CAD), especially in stable patients. This study assessed the ability
of an artificial intelligence algorithm (ECGio; HEARTio Inc, Pittsburgh, PA) to predict
the presence, location, and severity of coronary artery lesions in an unselected stable
patient population.
Methods
A cohort of 1659 stable outpatients was randomly divided into training (86%) and validation
(14%) subsets, maintaining population characteristics. ECGio was trained and validated
using electrocardiograms paired with retrospectively collected angiograms. Coronary
artery lesions were classified in 2 analyses. The primary classification was no to
mild (< 30% diameter stenosis [DS]) vs moderate (30%-70% DS) vs severe (≥ 70% DS)
CAD. The secondary classification was yes/no based on ≥ 50% DS in any vessel.
Results
In the primary analysis, 22 patients had no angiographic CAD and were grouped mild
CAD (56 patients, DS < 30%), 31 had moderate CAD (DS 30%-70%), and 113 had severe
CAD (DS ≥ 70%). Weighted average sensitivity was 93.2%, and weighted average specificity
was 96.4%. In the secondary analysis, 93 had significant CAD, and 128 did not. There
was sensitivity of 93.1% and specificity of 85.6% in determining the presence of clinically
significant disease (≥ 50% DS) in any vessel. ECGio was able to predict stenosis with
average vessel error in the left anterior descending coronary artery of 18%, the left
circumflex coronary artery of 19%, the right coronary artery of 18%, and the left
main coronary artery of 8%.
Conclusions
This study strongly suggests that it is possible to use an artificial intelligence
algorithm to determine the presence and severity of CAD in stable patients, using
data from a 12-lead electrocardiogram.
Résumé
Contexte
Actuellement, les algorithmes d'analyse d'un électrocardiogramme ne peuvent pas prédire
la présence d'une coronarienne (MC), en particulier chez les patients stables. Cette
étude visait à évaluer la capacité d'un algorithme d'intelligence artificielle (ECGio;
HEARTio Inc, Pittsburgh, PA) à prédire la présence, la localisation et la sévérité
des lésions coronariennes dans une population de patients stables non sélectionnés.
Méthodes
Une cohorte de 1659 patients stables, en consultation externe, a été divisée aléatoirement
en sous-catégories correspondant à l’apprentissage (86 %) et la validation (14 %),
en tenant compte des spécificités de la population. L'ECGio a été entraîné et validé
à l'aide d'électrocardiogrammes couplés à des angiogrammes recueillis rétrospectivement.
Les lésions des artères coronaires ont été classées dans deux analyses. La classification
primaire était identifiée comme MC nulle ou légère (sténose avec réduction du diamètre
[SD] < 30 %) vs modérée (SD 30 % -70 %) vs sévère (SD ≥ 70 %). La classification secondaire
était indexée comme oui/non sur la base d'une SD ≥ 50 % dans n'importe quel vaisseau.
Résultats
Concernant l'analyse principale, 22 patients n'avaient pas de MC à l'angiographie
et ont été regroupés en MC légère (56 patients, SD < 30 %), 31 avaient une MC modérée
(SD 30 % -70 %) et 113 avaient une MC sévère (DS ≥ 70 %). La sensibilité moyenne pondérée
était de 93,2 %, et la spécificité moyenne pondérée était de 96,4 %. Concernant l'analyse
secondaire, 93 personnes avaient une CM significative et 128 n'en avaient pas. La
sensibilité était de 93,1 % et la spécificité de 85,6 % lors de la détermination de
l'existence d'une pathologie cliniquement significative (SD ≥ 50 %) quel que soit
le vaisseau. L'ECGio a pu prédire la sténose avec une erreur moyenne, selon le vaisseau,
de 18 % dans l'artère coronaire antérieure descendante gauche, de 19 % dans l'artère
coronaire circonflexe gauche, de 18 % dans l'artère coronaire droite et de 8 % dans
l'artère coronaire principale gauche.
Conclusions
Cette étude suggère fortement qu'il est possible un algorithme d'intelligence artificielle
pour déterminer la présence et la gravité d'une MC chez des patients stables, à partir
des données d'un électrocardiogramme à 12 dérivations.
To read this article in full you will need to make a payment
Purchase one-time access:
Academic & Personal: 24 hour online accessCorporate R&D Professionals: 24 hour online accessOne-time access price info
- For academic or personal research use, select 'Academic and Personal'
- For corporate R&D use, select 'Corporate R&D Professionals'
Subscribe:
Subscribe to Canadian Journal of CardiologyAlready a print subscriber? Claim online access
Already an online subscriber? Sign in
Register: Create an account
Institutional Access: Sign in to ScienceDirect
References
- Diagnostic performance of electrocardiography in the assessment of significant coronary artery disease and its anatomical size in comparison with coronary angiography.J Res Med Sci. 2011; 16: 750
- Comprehensive strategy for the evaluation and triage of the chest pain patient.Ann Emerg Med. 1997; 29: 116-125
Ng A, Katanforoosh K. Splitting into Train, Dev and Test Sets Stanford University Computer Science. Available at: cs230.stanford.edu/blog/split/. Accessed August 21, 2020.
- PhysioBank, PhysioToolkit, and PhysioNet: components of a new research resource for complex physiologic signals.Circulation. 2000; 101: E215-E220
- The telemetric and Holter ECG warehouse initiative (THEW): a data repository for the design, implementation and validation of ECG-related technologies.Annu Int Conf IEEE Eng Med Biol Soc. 2010; 2010: 6252-6255
- Proposed Requirements for Cardiovascular Imaging-Related Machine Learning Evaluation (PRIME): a checklist. Reviewed by the American College of Cardiology Healthcare Innovation Council.JACC Cardiovasc Imaging. 2020; 13: 2017-2035
Shmueli B. Multi-class metrics made simple. Part I: precision and recall. Towards Data Science. Available at towardsdatascience.com/multi-class-metrics-made-simple-part-i-precision-and-recall-9250280bddc2. Accessed August 21, 2020.
- Diagnostic criteria for computer-aided electrocardiographic 15-lead system: evaluation using 12 leads and frank orthogonal leads with vector display.Heart. 1976; 38: 1247-1261
- Classification assessment methods.Appl Comput Informatics. 2021; 17: 168-192
Skelly AC, Hashimoto R, Buckley DI, et al. Noninvasive Testing for Coronary Artery Disease [Internet]. Rockville, MD: Agency for Healthcare Research and Quality (US); 2016. (Comparative Effectiveness Reviews, No. 171.) Available from https://www.ncbi.nlm.nih.gov/books/NBK361148. Accessed August 21, 2020.
- Prospective analysis of utility of signals from an ECG-enabled stethoscope to automatically detect a low ejection fraction using neural network techniques trained from the standard 12-lead ECG.Circulation. 2019; 140 (A13447)
- clinical prediction models for cardiovascular disease: Tufts predictive analytics and comparative effectiveness clinical prediction model database.Circ Cardiovasc Qual Outcomes. 2015; 8: 368-375
- Machine intelligence for management of acute coronary syndromes: neural or nervous times?.Can J Cardiol. 2020; 36: 470-473
- Artificial intelligence for diagnosis of acute coronary syndromes: a meta-analysis of machine learning approaches.Can J Cardiol. 2020; 36: 577-583
- Machine learning to predict stent restenosis based on daily demographic, clinical, and angiographic characteristics.Can J Cardiol. 2020; (36:1624–1632)
- Noninvasive Technologies for the Diagnosis of Coronary Artery Disease in Women. Comparative Effectiveness Review No. 58. (Prepared by the Duke Evidence-based Practice Center under Contract No. 290-2007-10066-I.) AHRQ Publication No. 12- EHC034-EF.Agency for Healthcare Research and Quality, Rockville, MD2012 (Accessed August 21, 2020)
- Frontline diagnostic evaluation of patients suspected of angina by coronary computed tomography reduces downstream resource utilization when compared to conventional ischemia testing.Int J Cardiovasc Imaging. 2011; 27: 813-823
- Diagnostic performance of stress myocardial perfusion imaging for coronary artery disease: a systematic review and meta-analysis.Eur Radiol. 2012; 22: 1881-1895
- Clinical implications of referral bias in the diagnostic performance of exercise testing for coronary artery disease.J Am Heart Assoc. 2013; 2 (:e000505)
- Does rubidium-82 PET have superior accuracy to SPECT perfusion imaging for the diagnosis of obstructive coronary disease? A systematic review and meta-analysis.J Am Coll Cardiol. 2012; 60: 1828-1837
- Diagnostic performance of noninvasive myocardial perfusion imaging using single photon emission computed tomography, cardiac magnetic resonance, and positron emission tomography imaging for the detection of obstructive coronary artery disease: a meta-analysis.J Am Coll Cardiol. 2012; 59: 1719-1728
- Diagnostic performance of stress cardiac magnetic resonance imaging in the detection of coronary artery disease: a meta-analysis.J Am Coll Cardiol. 2007; 50: 1343-1353
Skelly AC, Ecker ED, Henrikson NB, et al. Coronary Artery Calcium Scoring (CACS) as a Diagnostic Test for Detection of Coronary Artery Disease. Olympia, WA: Washington State Health Care Authority; Health Technology Assessment Program; September 4, 2009. https://www.hca.wa.gov/about-hca/health-technology-assessment/calcium-scoring. Accessed August 21, 2020.
- Diagnostic and prognostic value of absence of coronary artery calcification.JACC Cardiovasc Imaging. 2009; 2: 675-688
- Coronary CT angiography-derived fractional flow reserve: machine learning algorithm versus computational fluid dynamics modeling.Radiology. 2017; 285: 17-33
- Comparison of coronary computed tomography angiography, fractional flow reserve, and perfusion imaging for ischemia diagnosis.J Am Coll Cardiol. 2019; 73: 161-173
- Meta-analysis: diagnostic performance of low-radiation-dose coronary computed tomography angiography.Ann Intern Med. 2011; 154: 413-420
- A systematic review of the clinical effectiveness of 64-slice or higher computed tomography angiography as an alternative to invasive coronary angiography in the investigation of suspected coronary artery disease.BMC Cardiovasc Disord. 2011; 11: 32
Article info
Publication history
Published online: August 19, 2021
Accepted:
August 4,
2021
Received:
April 7,
2021
Footnotes
See editorial by Miller,pages 1691–1694of this issue.
See page 1723 for disclosure information.
Identification
Copyright
© 2021 Canadian Cardiovascular Society. Published by Elsevier Inc. All rights reserved.
ScienceDirect
Access this article on ScienceDirectLinked Article
- The Strength of a New SignalCanadian Journal of CardiologyVol. 37Issue 11
- PreviewWhereas Augustus Waller first recorded an “electrogram” from the intact skin’s surface in 1887, Willem Einthoven received the 1924 Nobel Prize in Medicine and Physiology for decoding intracardiac transmembrane anion transients into a clinically useful output in 1903.1 Einthoven’s Triangle for electrocardiogram (ECG) limb lead placement, designed to detect the electronic fields generated by a biological neural network during cyclical myocardial activity, is still used to amplify electrical potential changes between leads as dynamic vectors.
- Full-Text
- Preview
