跳到正文
原文
Nature 健康护理主题·· 2026-10-15AI 评分53

深度学习模型 DROID-RV 评估右心室大小与功能,与心血管结局相关

Deep learning-enabled assessment of right ventricular size and function is associated with cardiovascular outcomes

AI 导读

Mass General Brigham 与 Broad Institute 团队在 Nature Cardiovascular Research 发表研究,用基于超声心动图的深度学习模型 DROID-RV 评估右心室腔径与收缩功能,并分析其与房颤、心衰和死亡发生的关联。

正文

Data availability

MGB clinical and echocardiogram data contain protected health information and cannot be shared publicly. Limited, de-identified source data sufficient to recreate the presented figures are provided in the Supplementary Information. In addition, synthetic test data are available via GitHub at https://github.com/broadinstitute/ml4h/tree/master/model_zoo/DROID-RV to demonstrate the model training and inference pipelines. Reasonable requests for collaboration will be considered on a case-by-case basis by direct correspondence with the authors. RVENet data can be obtained for non-commercial research use by request from https://rvenet.github.io/dataset/.

Code availability

Model code and data processing scripts are available via GitHub at https://github.com/broadinstitute/ml4h/tree/master/model_zoo/DROID-RV. RVENet source code and data are available for request for non-commercial research use at https://rvenet.github.io/dataset/.

References

  1. Purmah, Y. et al. Right ventricular ejection fraction for the prediction of major adverse cardiovascular and heart failure-related events. Circ. Cardiovasc. Imaging 14, e011337 (2021).

    Article  PubMed  Google Scholar 

  2. Chatterjee, N. A. et al. Right ventricular structure and function are associated with incident atrial fibrillation. Circ. Arrhythm. Electrophysiol. 10, e004738 (2017).

    Article  PubMed  PubMed Central  Google Scholar 

  3. Palazzuoli, A. et al. Prognostic significance of an early echocardiographic evaluation of right ventricular dimension and function in acute heart failure. J. Card. Fail. 26, 813–820 (2020).

    Article  PubMed  Google Scholar 

  4. Hahn, R. T. et al. Multimodality imaging of right heart function. J. Am. Coll. Cardiol. 81, 1954–1973 (2023).

    Article  PubMed  Google Scholar 

  5. Lang, R. M. et al. Recommendations for cardiac chamber quantification by echocardiography in adults: an update from the American Society of Echocardiography and the European Association of Cardiovascular Imaging. J. Am. Soc. Echocardiogr. 28, 1–39.e14 (2015).

    Article  PubMed  Google Scholar 

  6. Rudski, L. G. et al. Guidelines for the echocardiographic assessment of the right heart in adults: a report from the American Society of Echocardiography. J. Am. Soc. Echocardiogr. 23, 685–713 (2010).

    Article  PubMed  Google Scholar 

  7. Lai, W. W. et al. Accuracy of guideline recommendations for two-dimensional quantification of the right ventricle by echocardiography. Int. J. Cardiovasc. Imaging 24, 691–698 (2008).

    Article  PubMed  Google Scholar 

  8. Ling, L. F. et al. Accuracy and interobserver concordance of echocardiographic assessment of right ventricular size and systolic function: a quality control exercise. J. Am. Soc. Echocardiogr. 25, 709–713 (2012).

    Article  PubMed  Google Scholar 

  9. Tromp, J. et al. Automated interpretation of systolic and diastolic function on the echocardiogram: a multicohort study. Lancet Digit. Health 4, e46–e54 (2022).

    Article  PubMed  CAS  Google Scholar 

  10. Ouyang, D. et al. Video-based AI for beat-to-beat assessment of cardiac function. Nature 580, 252–256 (2020).

    Article  PubMed  PubMed Central  CAS  Google Scholar 

  11. Ghorbani, A. et al. Deep learning interpretation of echocardiograms. npj Digit. Med. 3, 1–10 (2020).

    Article  Google Scholar 

  12. Lau, E. S. et al. Deep learning-enabled assessment of left heart structure and function predicts cardiovascular outcomes. J. Am. Coll. Cardiol. 82, 1936–1948 (2023).

    Article  PubMed  PubMed Central  Google Scholar 

  13. Al-Alusi, M. A. et al. A deep learning model to identify mitral valve prolapse from the echocardiogram. JACC Cardiovasc. Imaging 19, 18–29 (2026).

    Article  PubMed  Google Scholar 

  14. Tokodi, M. et al. Deep learning-based prediction of right ventricular ejection fraction using 2D echocardiograms. JACC Cardiovasc. Imaging 16, 1005–1018 (2023).

    Article  PubMed  Google Scholar 

  15. Beecy, A. N. et al. Development of novel machine learning model for right ventricular quantification on echocardiography—a multimodality validation study. Echocardiography 37, 688–697 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  16. Hagberg, E. et al. Semi-supervised learning with natural language processing for right ventricle classification in echocardiography—a scalable approach. Comput. Biol. Med. 143, 105282 (2022).

    Article  PubMed  Google Scholar 

  17. Kampaktsis, P. N. et al. An attention-based deep learning method for right ventricular quantification using 2D echocardiography: feasibility and accuracy. Echocardiography 41, e15719 https://doi.org/10.1111/echo.15719 (2024).

  18. Magyar, B. et al. RVENet: a large echocardiographic dataset for the deep learning-based assessment of right ventricular function. In Computer Vision–ECCV 2022 Workshops. Lecture Notes in Computer Science (eds Karlinsky, L. et al.) 569–583 (Springer, 2023).

  19. Singh, P. et al. One clinician is all you need—cardiac magnetic resonance imaging measurement extraction: deep learning algorithm development. JMIR Med. Inform. 10, e38178 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

  20. Khurshid, S. et al. Cohort design and natural language processing to reduce bias in electronic health records research. npj Digit. Med. 5, 47 (2022).

    Article  PubMed  PubMed Central  Google Scholar 

  21. von Elm, E. et al. Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies. BMJ 335, 806–808 (2007).

    Article  Google Scholar 

  22. Sengupta, P. P. et al. Proposed Requirements for Cardiovascular Imaging-Related Machine Learning Evaluation (PRIME): a checklist. JACC Cardiovasc. Imaging 13, 2017–2035 (2020).

    Article  PubMed  PubMed Central  Google Scholar 

  23. Khurshid, S., Keaney, J., Ellinor, P. T. & Lubitz, S. A. A simple and portable algorithm for identifying atrial fibrillation in the electronic medical record. Am. J. Cardiol. 117, 221–225 (2016).

    Article  PubMed  Google Scholar 

  24. Kondratyuk, D. et al. MoViNets: Mobile Video Networks for efficient video recognition. In 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 16015–16025 https://doi.org/10.1109/CVPR46437.2021.01576 (IEEE, 2021).

  25. Selvaraju, R. R. et al. Grad-CAM: visual explanations from deep networks via gradient-based localization. Int. J. Comput. Vis. 128, 336–359 (2020).

    Article  Google Scholar 

Download references

Acknowledgements

The authors would like to thank the creators of RVENet for sharing their dataset and model weights.

Funding

M.A.A. was supported by a grant from the National Institutes of Health (T32-HL007208) during this work. E.S.L. is supported by the NIH (K23HL159243 and R01HL180648), the American Heart Association, the Massachusetts Life Sciences Center, and the Tianqiao and Chrissy Chen Institute. S. Kany is supported by the Emmy Noether grant from the Deutsche Forschungsgemeinschaft (571960922). J.T.R. is supported by a Fellowship grant from the Sigrid Jusélius Foundation. S. Khurshid is supported by grants from the NIH (K23HL16989) and the AHA (23CDA1050571). J.E.H. is supported by grants from the NIH (R01 HL160003, R01 HL168889 and K24 HL153669). M.M. is supported by grants from the National Institutes of Health (3OT2OD035404-01S3, 1R01NS134597 and 1UG3HG014379-01) and from the American Heart Association (961045). S.A.L. was supported by NIH grants 1R01HL139731 and R01HL157635, and AHA 18SFRN34250007 during this work. P.T.E. is supported by grants from the National Institutes of Health (RO1HL092577, RO1HL157635 and RO1HL177209), from the American Heart Association (961045) and from the Foundation Leducq (24CVD01).

Author information

Authors and Affiliations

  1. Division of Cardiovascular Medicine, Mass General Brigham Heart and Vascular Institute, Boston, MA, USA

    Mostafa A. Al-Alusi, Emily S. Lau, Carl T. Andrews, Julian S. Haimovich, Shaan Khurshid, Michael H. Picard, Steven A. Lubitz & Patrick T. Ellinor

  2. Cardiovascular Research Center, Mass General Brigham Heart and Vascular Institute, Boston, MA, USA

    Mostafa A. Al-Alusi, Emily S. Lau, Julian S. Haimovich, Shaan Khurshid, Steven A. Lubitz & Patrick T. Ellinor

  3. Cardiovascular Disease Initiative, Broad Institute of MIT and Harvard, Cambridge, MA, USA

    Mostafa A. Al-Alusi, Kavya Kopparapu, Pulkit Singh, Paolo di Achille, Christopher Reeder, Emily S. Lau, Tal Shnitzer, Shinwan Kany, Joel T. Rämö, Julian S. Haimovich, Shaan Khurshid, Michael H. Picard, Jennifer E. Ho, Puneet Batra, Mahnaz Maddah, Steven A. Lubitz & Patrick T. Ellinor

  4. Harvard Medical School, Boston, MA, USA

    Mostafa A. Al-Alusi, Emily S. Lau, Julian S. Haimovich, Shaan Khurshid, Michael H. Picard, Jennifer E. Ho, Steven A. Lubitz & Patrick T. Ellinor

  5. Telemachus and Irene Demoulas Family Foundation Center for Cardiac Arrhythmias, Mass General Brigham Heart and Vascular Institute, Boston, MA, USA

    Mostafa A. Al-Alusi, Shaan Khurshid, Steven A. Lubitz & Patrick T. Ellinor

  6. Harvard John A. Paulson School of Engineering and Applied Sciences, Cambridge, MA, USA

    Kavya Kopparapu

  7. Machine Learning for Health, Broad Institute of MIT and Harvard, Cambridge, MA, USA

    Kavya Kopparapu, Pulkit Singh, Paolo di Achille, Christopher Reeder, Tal Shnitzer, Puneet Batra & Mahnaz Maddah

  8. Department of Cardiology, University Heart and Vascular Center Hamburg-Eppendorf, Hamburg, Germany

    Shinwan Kany

  9. Partner Site North, German Center for Cardiovascular Research (DZHK), Hamburg, Germany

    Shinwan Kany

  10. Institute for Molecular Medicine Finland, Helsinki Institute of Life Science, University of Helsinki, Helsinki, Finland

    Joel T. Rämö

  11. Cardiovascular Institute and Division of Cardiology, Department of Medicine, Beth Israel Deaconess Medical Center, Boston, MA, USA

    Jennifer E. Ho

Authors

  1. Mostafa A. Al-Alusi
  2. Kavya Kopparapu
  3. Pulkit Singh
  4. Paolo di Achille
  5. Christopher Reeder
  6. Emily S. Lau
  7. Tal Shnitzer
  8. Carl T. Andrews
  9. Shinwan Kany
  10. Joel T. Rämö
  11. Julian S. Haimovich
  12. Shaan Khurshid
  13. Michael H. Picard
  14. Jennifer E. Ho
  15. Puneet Batra
  16. Mahnaz Maddah
  17. Steven A. Lubitz
  18. Patrick T. Ellinor

Contributions

M.A.A. led the study and contributed to study design, data curation, cohort development, model development, statistical analyses, initial paper drafting and paper revision. K.K., P.S. and P.d.A. were the technical leads for the study and contributed to study design, data curation, cohort development, model design and development, and statistical analyses. C.R. contributed to data curation, cohort development and model design. E.S.L. contributed to study design, data curation and clinical oversight for model development. T.S. contributed to model development. C.T.A. provided curated echocardiogram data. S. Kany and J.T.R. contributed to study design and paper revision. J.S.H. contributed to data curation and cohort development. S. Khurshid contributed to study design, data curation, cohort development and paper revision. M.H.P. and J.E.H. supervised study design and contributed to paper revision. P.B. and M.M. supervised model design and development, contributed to paper revision and provided funding. S.A.L. supervised study design and cohort development, provided funding and contributed to paper revision. P.T.E. was the primary clinical supervisor for the study, including supervision of study design, paper revision and providing funding.

Corresponding author

Correspondence to Patrick T. Ellinor.

Ethics declarations

Competing interests

P.T.E., E.S.L., J.E.H., M.A.A. and S.A.L. are inventors on a patent (US 12,357,278 B2) related to this work issued on July 15, 2025. M.A.A. reports consulting for GLG, Inc., unrelated to this work. E.S.L. has served on advisory boards or consulted for Bayer AG, SystoleHealth, Inc., Amissa Health and Roon, unrelated to this work. P.d.A. is a full-time employee of Google as of June 2022. K.K. is a full-time employee of DeepMind as of August 2022. M.H.P. has consulted for Vertex Pharmaceuticals, unrelated to this work. J.E.H. has previously received sponsored research support from Bayer AG. P.B. is an employee of Ember AI, has received sponsored research support from Bayer AG and IBM, and has consulted for Novartis and Prometheus Biosciences, unrelated to this work. S.A.L. is employed by Novartis as of July 2022; has received sponsored research support from Bristol Myers Squibb, Pfizer, Boehringer Ingelheim, Fitbit, Medtronic, Premier and IBM; and has consulted for Bristol Myers Squibb, Pfizer, Blackstone Life Sciences and Invitae, unrelated to this work. P.T.E. receives sponsored research support from Bayer AG, Bristol Myers Squibb, Pfizer and Novo Nordisk; he has also served on advisory boards or consulted for Bayer AG and is a member of the Board of Directors of Editas Medicine, Inc., unrelated to this work. The other authors declare no competing interests.

Peer review

Peer review information

Nature Cardiovascular Research thanks Bjørnar Leangen Grenne, Steffen Petersen and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Extended data

Extended Data Fig. 1 Model Architecture.

DROID-RV is built on a MoViNet backbone, which is cut off and flattened at the final convolutional layer. This layer is connected to a dense layer with 256 nodes, which is connected to 3 classification heads (sex [‘cls_SexDSC_clean’], right ventricular systolic function [‘cls_RV_SystolicFunction_clean’], and right ventricular cavity size [‘cls_RV_CavitySize_clean’]) and 1 regression head (‘echolab’) with 2 outputs (age and right ventricular end-diastolic diameter). Dropout layers are present between dense layer connections. During pre-training an identical architecture is used with only 1 output in the ‘echolab’ regression head for age.

Source data

Source Data Figs. 2–4 (download XLSX )

Source Data 2: Sheet 1 includes predicted and echocardiographer-measured RVEDD and sex for the internal and external test sets. Source Data 3: Sheet 1 includes predicted RV dysfunction probability and echocardiographer-assessed categorical RV function for the internal and external test sets. Sheet 2 includes predicted and measured RVEF for the RVENet test sample. Source Data 4: Sheets 2–6 include survival time in years, censor/event status and RV size/function group for incidence of atrial fibrillation, heart failure and mortality.

Rights and permissions

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Al-Alusi, M.A., Kopparapu, K., Singh, P. et al. Deep learning-enabled assessment of right ventricular size and function is associated with cardiovascular outcomes. Nat Cardiovasc Res (2026). https://doi.org/10.1038/s44161-026-00886-4

Download citation

  • Received: 23 December 2024

  • Accepted: 08 September 2026

  • Published: 15 October 2026

  • Version of record: 15 October 2026

  • DOI: https://doi.org/10.1038/s44161-026-00886-4

来源:Nature 健康护理主题 · nature.com