DeepFake electrocardiograms using generative adversarial networks are the beginning of the end for privacy issues in medicine

Recent global developments underscore the prominent role big data have in modern medical science. But privacy issues constitute a prevalent problem for collecting and sharing data between researchers. However, synthetic data generated to represent real data carrying similar information and distribution may alleviate the privacy issue. In this study, we present generative adversarial networks (GANs) capable of generating realistic synthetic DeepFake 10-s 12-lead electrocardiograms (ECGs). We have developed and compared two methods, named WaveGAN* and Pulse2Pulse. We trained the GANs with 7,233 real normal ECGs to produce 121,977 DeepFake normal ECGs. By verifying the ECGs using a commercial ECG interpretation program (MUSE 12SL, GE Healthcare), we demonstrate that the Pulse2Pulse GAN was superior to the WaveGAN* to produce realistic ECGs. ECG intervals and amplitudes were similar between the DeepFake and real ECGs. Although these synthetic ECGs mimic the dataset used for creation, the ECGs are not linked to any individuals and may thus be used freely. The synthetic dataset will be available as open access for researchers at OSF.io and the DeepFake generator available at the Python Package Index (PyPI) for generating synthetic ECGs. In conclusion, we were able to generate realistic synthetic ECGs using generative adversarial neural networks on normal ECGs from two population studies, thereby addressing the relevant privacy issues in medical datasets.

Deep neural network-estimated electrocardiographic age as a mortality predictor

The electrocardiogram (ECG) is the most commonly used exam for the evaluation of cardiovascular diseases. Here we propose that the age predicted by artificial intelligence (AI) from the raw ECG (ECG-age) can be a measure of cardiovascular health. A deep neural network is trained to predict a patient’s age from the 12-lead ECG in the CODE study cohort (n = 1,558,415 patients). On a 15% hold-out split, patients with ECG-age more than 8 years greater than the chronological age have a higher mortality rate (hazard ratio (HR) 1.79, p < 0.001), whereas those with ECG-age more than 8 years smaller, have a lower mortality rate (HR 0.78, p < 0.001). Similar results are obtained in the external cohorts ELSA-Brasil (n = 14,236) and SaMi-Trop (n = 1,631). Moreover, even for apparent normal ECGs, the predicted ECG-age gap from the chronological age remains a statistically significant risk predictor. These results show that the AI-enabled analysis of the ECG can add prognostic information.

Development of AI-based method to detect the subtle ECG deviations from the population ECG norm

Funding Acknowledgements Type of funding sources: Public grant(s) – National budget only. Main funding source(s): National Academy of Science of Ukraine Background Electrocardiogram (ECG) is still the primary source for the diagnostic and prognostic information about cardiovascular diseases. The concept of "normal ECG" parameters is crucial for the reliable diagnosis, since it provides reference for the ECG under examination. With the development of new methods and tools for ECG feature extraction and classification based on artificial intelligence (AI), it becomes possible to identify subtle changes in the heart activity to detect possible abnormalities at the early stage. The challenge of this work is to identify the deviations in ECG of clinically healthy persons from the conditional "population" norm . Methods The normal ECG is described as a feature vector composed of the time-magnitude parameters of signal-averaged ECG (SAECG). To define the subjects that possibly have variations from the "population" norm, the outlier detection approach is proposed: first the cloud of the vectors , constructed from the set of normal ECG"s , obtained from young, clinically similar healthy persons was created in feature space. Then, a particular ECG is considered deviant and requires the attention of the clinician when it is considered an outlier of the cloud of normal ECGs. In the experiment, SAECGs from the group of 139 young subjects (male, age 18-28 years) with no reported cardiovascular problems are used to extract 34 features from SAECG leads (magnitudes and durations of ECG waves, duration of ECG segments, etc.). ECGs were routinely previewed by qualified physicians, and no obvious anomalies were noticed. The Isolation Forest anomaly detection method is used with variable numbers of trees and different contamination parameters. Results The ratio of outliers were changed from 5 to 10% (7-12 subjects) with various numbers of estimator trees. Seven outlier SAECGs were repeatedly appearing for various settings. Out of these, 4 subjects were the oldest persons in group examined , and 3 others had a rare ventricular premature beats during routine ECG examination. Conclusion The proposed method is promising for application in routine and express ECG tests since it is able to quantify the subtle deviations from the normal ECG group.

DeepFake electrocardiograms: the beginning of the end for privacy issues in medicine

SummaryBig data is needed to implement personalized medicine, but privacy issues are a prevalent problem for collecting data and sharing them between researchers. A solution is synthetic data generated to represent real dataset carrying similar information.Here, we present generative adversarial networks (GANs) capable of generating realistic synthetic DeepFake 12-lead 10-sec electrocardiograms (ECGs). We have developed and compare two methods, namely WaveGAN* and Pulse2Pulse GAN. We trained the GANs with 7,233 real normal ECG to produce 121,977 DeepFake normal ECGs. By verifying the ECGs using a commercial ECG interpretation program (MUSE 12SL, GE Healthcare), we demonstrate that the Pulse2Pulse GAN was superior to the WaveGAN to produce realistic ECGs. ECG intervals and amplitudes were similar between the DeepFake and real ECGs. These synthetic ECGs are fully anonymous and cannot be referred to any individual, hence they may be used freely. The synthetic dataset will be available as open access for researchers at OSF.io and the DeepFake generator available at the Python Package Index (PyPI) for generating synthetic ECGs.In conclusion, we were able to generate realistic synthetic ECGs using adversarial neural networks on normal ECGs from two population studies, i.e., there by solving the relevant privacy issues in medical datasets.

Deep neural network estimated electrocardiographic-age as a mortality predictor

The electrocardiogram (ECG) is the most commonly used exam for the screening and evaluation of cardiovascular diseases. Here we propose that the age predicted by artificial intelligence (AI) from the raw ECG tracing (ECG-age) can be a measure of cardiovascular health and provide prognostic information. A deep convolutional neural network was trained to predict a patient’s age from the 12-lead ECG using data from patients that underwent an ECG from 2010 to 2017 - the CODE study cohort (n=1,558,415 patients). On the 15% hold-out CODE test split, patients with ECG-age more than 8 years greater than chronological age had a higher mortality rate (hazard ratio (HR) 1.79, p<0.001) in a mean follow-up of 3.67 years, whereas those with ECG-age more than 8 years less than chronological age had a lower mortality rate (HR 0.78, p<0.001). Similar results were obtained in the external cohorts ELSA-Brasil (n=14,236) and SaMi-Trop (n=1,631). The ability to predict mortality from the ECG predicted age remains even when we adjust the model for cardiovascular risk factors. Moreover, even for apparent normal ECGs, having a predicted ECG-age 8 or more years greater than chronological age remained a statistically significant predictor of risk (HR 1.53, p<0.001 in CODE 15% test split). These results show that AI-enabled analysis of the ECG can add prognostic information to the interpretation of the 12-lead ECGs.

Prospective associations between ECG abnormalities and death or myocardial infarction in a cohort of 980 employed, middle-aged Swedish men

Background In 1993, 1000 randomly selected employed Swedish men aged 45–50 years were invited to a nurse-led health examination with a survey on life style, fasting lab tests, and a 12-lead ECG. A repeat examination was offered in 1998. The ECGs were classified according to the Minnesota Code. Upon ethical approval, endpoints in terms of MI and death over 25 years were collected from Swedish national registers with the purpose of analyzing the independent association of ECG abnormalities as risk factors for myocardial infarction and death. Results Seventy-nine of 977 participants had at least one ECG abnormality 1993 or 1998. One hundred participants had a first MI over the 25 years. Odds ratio for having an MI in the group that had one or more ECG abnormality compared with the group with two normal ECGs was estimated to 3.16. 95%CI (1.74; 5.73), p value 0.0001. One hundred fifty-seven participants had died before 2019. For death, similarly no statistically significant difference was shown, OR 1.52, 95%CI (0.83; 2.76). Conclusions Our study suggests that presence of ST- and R-wave changes is associated with an independent 3–4-fold increased risk of MI after 25 years follow-up, but not of death. A 12-lead resting ECG should be included in any MI risk calculation on an individual level.

P1496 Congenital anomalous origin of the left coronary system

A 62 year old gentleman presented with a history of recurrent central chest pains radiating to the left arm and jaw lasting up to 15-20 minutes and relieved with GTN. He had numerous admissions to hospital over a period of three years with negative Troponins and normal ECGs. There were several cardiovascular risk factors including obesity, diabetes, hypertension and dyslipidaemia. He also had a family history of ischaemic heart disease, with his mother and brother having heart attacks in their 60s. He was referred for outpatient investigations on multiple occasions but did not attend. This lead to a delay in a formal diagnosis until we eventually convinced him to undergo invasive diagnostic coronary angiography in June 2018. Prior to this, an Echocardiogram was done and showed reasonably preserved cardiac systolic function. Coronary angiography demonstrated unique anatomical distribution of the three main coronary vessels, with an anomalous origin of the left main system (LMS) and left sided arteries arising from the right coronary cusp. The right coronary artery stemmed from its natural position and was the dominant vessel. Hence, all the coronary arteries arose from the same cusp. The LMS was anomalous and hypoplastic; an exceedingly rare occurrence of less than 0.03%. These unusual findings were then confirmed on CT Coronary Angiogram. Although a surgical opinion was sought, the decision was a non-operative approach in view of no significant obstructive lesions and given the technical difficulties of undertaking coronary bypass. Viability imaging and ischaemia testing were then pursued with nuclear modalities. Ultimately, it was proven that the lesions did not show any significant reversible ischaemia and so a continued aggressive secondary prevention strategy was adopted. The patient is stable and doing well on optimal medical therapy. Abstract P1496 Figure. LMS Arising From Right Coronary Cusp

5105A deep neural network predicts atrial fibrillation from normal ECGs recorded on a smartphone-enabled device

Background Electrocardiographic predictors of atrial fibrillation (AF) from a non-AF ECG–such as p wave abnormalities and supraventricular ectopy–have been extensively documented. However, risk prediction tools for AF utilize little if any of the wealth of information available from the ECG. Better AF prediction from the ECG may improve efficiency of screening and performance of AF risk tools. Deep learning methods have the potential to extract an unlimited number of features from the ECG to improve prediction of AF. Purpose We hypothesize that a deep learning model can identify patterns predictive of AF during normal sinus rhythm. To test the hypothesis, we trained and tested a neural network to predict AF from normal sinus rhythm ambulatory ECG data. Methods We trained a deep convolutional neural network to detect features of AF that are present in single-lead ECGs with normal sinus rhythm, recorded using a Food and Drug Administration (FDA)-cleared, smartphone-enabled device. A cohort of 27,526 patients with at least 50 ECGs recorded between January 7, 2013, and September, 19, 2018, and the FDA-cleared automated findings of Normal and Atrial Fibrillation associated with those ECGs, were used for model development. Specifically, we trained the deep learning model on 1,984,581 Normal ECGs from 19,267 patients with 1) only Normal ECG recordings, or 2) at least 30% ECGs with AF. Of the 27,526 patients, an internal set of 8,259 patients with 841,776 Normal ECGs was saved for testing (validation). Results Among 8,259 patients in the test set, 3,467 patients had at least 30% of their ECGs with an automated finding of AF. When the deep learning model was run on 841,776 Normal ECGs, it was able to predict whether the ECG was from a patient with no AF or with 30% or more AF, with an area under the curve (AUC) of 0.80. Using an operating point with equal sensitivity and specificity, the model's sensitivity and specificity were 73.1%. Using an operating point with high specificity (90.0%), the model's sensitivity was 48.0%. When the model was applied to a randomly-selected, broader cohort of 15,000 patients (at least 50 ECGs recorded, any amount of AF), a positive, non-linear relationship between neural network output and AF burden per patient was observed (Figure). Model Output vs AF Burden Per Patient Conclusions A deep learning model was able to predict AF from ECGs in normal sinus rhythm that were recorded on a smartphone-enabled device. The use of deep learning, if prospectively validated, may facilitate AF screening in patients with paroxysmal disease or warn patients who are at high risk for developing AF. Acknowledgement/Funding AliveCor

Out-of-hospital cardiac arrest due to idiopathic ventricular fibrillation in patients with normal electrocardiograms: results from a multicentre long-term registry

Aims To define the clinical characteristics and long-term clinical outcomes of a large cohort of patients with idiopathic ventricular fibrillation (IVF) and normal 12-lead electrocardiograms (ECGs). Methods and results Patients with ventricular fibrillation as the presenting rhythm, normal baseline, and follow-up ECGs with no signs of cardiac channelopathy including early repolarization or atrioventricular conduction abnormalities, and without structural heart disease were included in a registry. A total of 245 patients (median age: 38 years; males 59%) were recruited from 25 centres. An implantable cardioverter-defibrillator (ICD) was implanted in 226 patients (92%), while 18 patients (8%) were treated with drug therapy only. Over a median follow-up of 63 months (interquartile range: 25–110 months), 12 patients died (5%); in four of them (1.6%) the lethal event was of cardiac origin. Patients treated with antiarrhythmic drugs only had a higher rate of cardiovascular death compared to patients who received an ICD (16% vs. 0.4%, P = 0.001). Fifty-two patients (21%) experienced an arrhythmic recurrence. Age ≤16 years at the time of the first ventricular arrhythmia was the only predictor of arrhythmic recurrence on multivariable analysis [hazard ratio (HR) 0.41, 95% confidence interval (CI) 0.18–0.92; P = 0.03]. Conclusion Patients with IVF and persistently normal ECGs frequently have arrhythmic recurrences, but a good prognosis when treated with an ICD. Children are a category of IVF patients at higher risk of arrhythmic recurrences.