The electrocardiogram (ECG) is arguably the cardiologist’s best friend. Willem Einthoven’s invention created one of the most widely used diagnostic tests in clinical practice. The ECG is an easily available, patient-friendly, noninvasive, inexpensive, and reproducible technique, without peer for the diagnosis of myocardial ischemia, cardiac arrhythmias, structural changes of the myocardium, drug effects, and electrolyte and metabolic disturbances.1 In addition to this, the ECG can provide information about the extent and severity of ischemia in acute coronary syndromes, assist in the localization of the site or pathway for tachycardias, identify heart failure patients who would benefit from cardiac resynchronization and identify familial diseases with risk of sudden cardiac death.
The ECG detects pathological changes prior to the development of structural changes in the heart. For instance, a strain pattern, defined as a down-sloping convex ST segment with inverted asymmetrical T-wave opposite the QRS axis in lead V5 or V6, is predictive of future risk of HF and death in hypertensive patients.2 In fact, the 12 lead surface ECG is but one format to represent the electrical activity of the heart. Small changes in the morphology of the surface ECG, not visible to the human eye may reflect significant shifts in electrochemical messaging. Techniques such as signal averaged ECG, vectorcardiography aim to overcome these limitations and have been around for several decades.3 However, sophisticated and more advanced applications of the ECG have not found their way into the routine practice of clinical cardiology. Most of these are limited by low sensitivity which prohibits widespread application.
With advances in computational techniques and availability of big data collected from a variety of sources, there have been advances in unpacking the information encoded in several biologic signals. Deep learning is a type of machine learning technique with a diverse set of validated applications such as facial and speech recognition. Using deep learning to analyze retinal fundus images, Google has developed an algorithm that makes a diagnosis of diabetic retinopathy with a high degree of accuracy comparable to ophthalmologists.4 Furthermore, this algorithm detects cardiovascular events and even identifies gender from retinal images alone.5 This work represents a new way of scientific discovery, an alternative to the traditional hypothesis driven research approach. A data-driven approach can help generate newer hypothesis-guided experiments.
Deep learning for the analysis of ECG signals is an area of active research. Hannun and colleagues have shown comparable accuracy and even higher sensitivity for classifying arrhythmias using a deep neural network model versus board certified cardiologists.6 Work presented at scientific sessions 2019 from the University of Dusseldorf by Makimoto and colleagues showed that a convolutional neural network (CNN) was able to diagnose myocardial infarctions (MI) with more accuracy than cardiologists.7 The accuracy of MI recognition in ECGs by CNN was 84±2%, which was significantly higher than by cardiologists (64±7%, p<0.001). Designing clinical workflows where deep learning models provide rapid, expert level over read of ECGs, complemented by human oversight can have significant clinical impact.
The ECG signal represents the various electrical, chemical and mechanical events during the cardiac cycle. Deep learning algorithms have been able to decode these signals to make predictions about LVEF and diastolic dysfunction based on ECG data alone. Sengupta and colleagues used continuous wavelet transformation for post-processing the ECG signals and correlated several derived features for predicting abnormal myocardial relaxation as defined by abnormal tissue doppler.8 The area under the curve for their machine learning model for prediction of abnormal myocardial mechanical relaxation was 91% [CI: 0.86-0.95]. Attia and colleagues from Mayo Clinic presented their work at scientific sessions 2019 on predicting LVEF using single lead ECG signals acquired by an ECG-enabled stethoscope.9 A neural network previously used on 12-lead ECG for predicting EF was trained on single lead ECG data and was able to predict low EF with an area under the curve of 0.88 [CI:0.80-0.94] for EF<=35% and 0.81 [CI:0.72-0.88] for EF<50%
These findings and many others are constantly expanding the utility of the ECG in clinical practice wherein, the ECG provides more nuanced and finer details such as ejection fraction and predicts future outcomes with high accuracy. This represents significant progress from the days of the string galvanometer of Einthoven. Modern cardiovascular medicine is faced with many challenges related to prevention, diagnosis and treatment of disease, compounded by rising healthcare costs. The inexpensive and reliable best friend of the cardiologist – the ECG – can reveal its secrets to tackle these problems. The words of Einthoven remind us that there remains much to be done for decoding these secrets, “An instrument takes its true value not so much from the work it possibly might do but from the work it really does.”10 Future research is needed to validate the promise of these exciting new findings.
- Wellens HJ, Gorgels AP. The electrocardiogram 102 years after einthoven. Circulation. 2004;109:562-564
- Okin PM, Devereux RB, Nieminen MS, Jern S, Oikarinen L, Viitasalo M, Toivonen L, Kjeldsen SE, Dahlof B, Investigators LS. Electrocardiographic strain pattern and prediction of new-onset congestive heart failure in hypertensive patients: The losartan intervention for endpoint reduction in hypertension (life) study. Circulation. 2006;113:67-73
- Gatzoulis KA, Arsenos P, Trachanas K, Dilaveris P, Antoniou C, Tsiachris D, Sideris S, Kolettis TM, Tousoulis D. Signal-averaged electrocardiography: Past, present, and future. J Arrhythm. 2018;34:222-229
- Gulshan V, Peng L, Coram M, Stumpe MC, Wu D, Narayanaswamy A, Venugopalan S, Widner K, Madams T, Cuadros J, Kim R, Raman R, Nelson PC, Mega JL, Webster DR. Development and validation of a deep learning algorithm for detection of diabetic retinopathy in retinal fundus photographs. JAMA. 2016;316:2402-2410
- Poplin R, Varadarajan AV, Blumer K, Liu Y, McConnell MV, Corrado GS, Peng L, Webster DR. Prediction of cardiovascular risk factors from retinal fundus photographs via deep learning. Nat Biomed Eng. 2018;2:158-164
- Hannun AY, Rajpurkar P, Haghpanahi M, Tison GH, Bourn C, Turakhia MP, Ng AY. Cardiologist-level arrhythmia detection and classification in ambulatory electrocardiograms using a deep neural network. Nat. Med. 2019;25:65-+
- Makimoto H, Hoeckmann M, Gerguri S, Clasen L, Schmidt J, Assadi-Schmidt A, Bejinariu A, Mueller P, Gloeckner D, Angendohr S, Brinkmeyer C, Kelm M. Abstract 13914: Artificial intelligence finds myocardial infaction in ecg more accurately than cardiologists. Circulation. 2019;140:A13914-A13914
- Sengupta PP, Kulkarni H, Narula J. Prediction of abnormal myocardial relaxation from signal processed surface ecg. J Am Coll Cardiol. 2018;71:1650-1660
- Attia ZI, Dugan J, Maidens J, Rideout A, Lopez-Jimenez F, Noseworthy PA, Asirvatham S, Pellikka PA, Ladewig DJ, Satam G, Pham S, Venkatraman S, Friedman P, Kapa S. Abstract 13447: 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-A13447
- Rosen MR. The electrocardiogram 100 years later: Electrical insights into molecular messages. Circulation. 2002;106:2173-2179
The views, opinions and positions expressed within this blog are those of the author(s) alone and do not represent those of the American Heart Association. The accuracy, completeness and validity of any statements made within this article are not guaranteed. We accept no liability for any errors, omissions or representations. The copyright of this content belongs to the author and any liability with regards to infringement of intellectual property rights remains with them. The Early Career Voice blog is not intended to provide medical advice or treatment. Only your healthcare provider can provide that. The American Heart Association recommends that you consult your healthcare provider regarding your personal health matters. If you think you are having a heart attack, stroke or another emergency, please call 911 immediately.
Sumeet Pawar, MD is a cardiology fellow at Yale University and interested in healthcare innovation, digital health and leadership. He is currently working on piloting a digital platform for streamlining hospital operations by democratizing access to inpatient clinical data for patients and their families. Follow on Twitter: @sumeetpawar