Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2023 Feb 14;3(1):24.
doi: 10.1038/s43856-023-00240-w.

Multi-center retrospective cohort study applying deep learning to electrocardiograms to identify left heart valvular dysfunction

Akhil Vaid  1   2 , Edgar Argulian  3   4 , Stamatios Lerakis  3   4 , Brett K Beaulieu-Jones  5   6 , Chayakrit Krittanawong  7 , Eyal Klang  8   9 , Joshua Lampert  10 , Vivek Y Reddy  10 , Jagat Narula  3   4   11 , Girish N Nadkarni  12   13 , Benjamin S Glicksberg  14   15
Affiliations

Multi-center retrospective cohort study applying deep learning to electrocardiograms to identify left heart valvular dysfunction

Akhil Vaid et al. Commun Med (Lond). .

Abstract

Background: Aortic Stenosis and Mitral Regurgitation are common valvular conditions representing a hidden burden of disease within the population. The aim of this study was to develop and validate deep learning-based screening and diagnostic tools that can help guide clinical decision making.

Methods: In this multi-center retrospective cohort study, we acquired Transthoracic Echocardiogram reports from five Mount Sinai hospitals within New York City representing a demographically diverse cohort of patients. We developed a Natural Language Processing pipeline to extract ground-truth labels about valvular status and paired these to Electrocardiograms (ECGs). We developed and externally validated deep learning models capable of detecting valvular disease, in addition to considering scenarios of clinical deployment.

Results: We use 617,338 ECGs paired to transthoracic echocardiograms from 123,096 patients to develop a deep learning model for detection of Mitral Regurgitation. Area Under Receiver Operating Characteristic curve (AUROC) is 0.88 (95% CI:0.88-0.89) in internal testing, and 0.81 (95% CI:0.80-0.82) in external validation. To develop a model for detection of Aortic Stenosis, we use 617,338 Echo-ECG pairs for 128,628 patients. AUROC is 0.89 (95% CI: 0.88-0.89) in internal testing, going to 0.86 (95% CI: 0.85-0.87) in external validation. The model's performance increases leading up to the time of the diagnostic echo, and it performs well in validation against requirement of Transcatheter Aortic Valve Replacement procedures.

Conclusions: Deep learning based tools can increase the amount of information extracted from ubiquitous investigations such as the ECG. Such tools are inexpensive, can help in earlier disease detection, and potentially improve prognosis.

Plain language summary

The valves of the heart have flaps that open and close when the heart beats to maintain the flow of blood in the correct direction. Valvular disease, such as backflow or narrowing, puts additional strain upon heart muscles which can lead to heart failure. Usually, these conditions are diagnosed by doing an echocardiogram, an ultrasound scan of the heart and nearby blood vessels. The electrocardiogram (ECG) records the electrical signal generated by the heart and can be obtained more easily. We used deep learning neural networks, self-learning computer algorithms which excel at finding patterns within complex data. This enabled us to develop computer software able to diagnose valvular disease from ECGs. Earlier detection of such disease can help in improving overall outcome, while also reducing costs related to treatment.

PubMed Disclaimer

Conflict of interest statement

B.S.G. has received consulting fees from Anthem AI and consulting and advisory fees from Prometheus Biosciences. G.N.N. has received consulting fees from AstraZeneca, Reata, BioVie, Siemens Healthineers and GLG Consulting; grant funding from Goldfinch Bio and Renalytix; financial compensation as a scientific board member and adviser to Renalytix; owns equity in Renalytix and Pensieve Health as a cofounder and is on the advisory board of Neurona Health. The other authors declare no competing interests

Figures

Fig. 1
Fig. 1. Flow diagram showing numbers of patients and paired ECG investigations at each step of data preprocessing.
Numbers indicate investigations following initial data collection, followed by parsing echo reports for relevant diagnostic terms, temporal restriction, and removal of outliers based on mathematical analysis of waveforms.
Fig. 2
Fig. 2. Receiver Operating Characteristic (ROC) Curves.
Panel a Mitral Regurgitation. Panel b Aortic Stenosis. Area Under Curve (95% Confidence Interval) with shaded area around curve representing confidence interval. Red dashed line represents floor of performance as in the case of a hypothetical model making purely random predictions. Overall dataset size: 607,429 Echo-ECG pairs for 123,096 patients for Mitral Regurgitation. 617,338 Echo-ECG pairs for 128,628 patients for Aortic Stenosis.
Fig. 3
Fig. 3. Model performance by age, sex, and race subgroups.
Panel a Mitral Regurgitation. Panel b Aortic Stenosis. Values presented are Area Under Receiver Operating Characteristic Curve (AUROC). Bar segments at bottom delineate internal testing and external validation by color. Inner circles represent groups by age/US Census defined racial categories. Source data for figure is available in Supplementary Data 1–6.
Fig. 4
Fig. 4. Model interpretability: Mitral Regurgitation.
Panel a Input pixels most responsible for driving the prediction towards the outcome are highlighted. Panel b Relative contributions of waveform / tabular data to the final prediction. Panel c Relative importance of tabular features with respect to each other. Patient (n = 1) was positive for Mitral Regurgitation. Source data for figure is available in Supplementary Data 7.
Fig. 5
Fig. 5. Model interpretability: Aortic Stenosis.
Panel a Input pixels most responsible for driving the prediction towards the outcome are highlighted. Panel b Relative contributions of waveform / tabular data to the final prediction. Panel c Relative importance of tabular features with respect to each other. Patient (n = 1) was positive for Aortic Stenosis. Source data for figure is available in Supplementary Data 7.
Fig. 6
Fig. 6. Cumulative incidence of Transcatheter Aortic Valve Replacement (TAVR) by model prediction.
At risk numbers represent multiple predictions paired to each TAVR procedure for each patient prior to the date of the procedure. Shaded area around curve represents confidence interval. Follow up interval: 5 years.
Fig. 7
Fig. 7. Model performance at detection of Aortic Stenosis prior to diagnostic echo.
Shaded area around curve represents confidence interval. Highlighted points on each curve demonstrate optimal sensitivity and specificity as derived by the Youden J. Panel a n = 2062 patients (3–6 months), Panel c 2096 patients (6–12 months), Panel b 2076 patients (12–18 months) and Panel d 2058 patients (18–24 months) AUROC: Area Under Receiver Operating Characteristic Curve.
See this image and copyright information in PMC

References

    1. Iung B, Vahanian A. Epidemiology of valvular heart disease in the adult. Nat. Rev. Cardiol. 2011;8:162–172. doi: 10.1038/nrcardio.2010.202. - DOI - PubMed
    1. Montant P, et al. Long-term survival in asymptomatic patients with severe degenerative mitral regurgitation: a propensity score-based comparison between an early surgical strategy and a conservative treatment approach. J. Thorac. Cardiovasc. Surg. 2009;138:1339–1348. doi: 10.1016/j.jtcvs.2009.03.046. - DOI - PubMed
    1. Travis B, Partho PS, Jagat N. Is TAVR ready for the global aging population? Global Heart. 2017;12:291–299. doi: 10.1016/j.gheart.2017.02.002. - DOI - PubMed
    1. Kundi H, et al. Trends in isolated surgical aortic valve replacement according to hospital-based transcatheter aortic valve replacement volumes. JACC Cardiovasc. Interv. 2018;11:2148–2156. doi: 10.1016/j.jcin.2018.07.002. - DOI - PubMed
    1. Coleman W, Weidman-Evans E, Clawson R. Diagnosing and managing mitral regurgitation. J. Am. Acad. of PAs. 2017;30:11–14. - PubMed