More than 300 million electrocardiograms (ECGs) are obtained annually worldwide and the correct interpretation of the ECG is pivotal for accurate diagnosis of many cardiac abnormalities. Both physician and computerized interpretation of the ECG have not been able to reach cardiologist level accuracy in detecting (acute) cardiac abnormalities. The most promising development in the time of artificial intelligence and big data is the use of integrated self-learning algorithms, called deep neural networks. We propose to use deep learning for automated ECG interpretation in the fields where it is most unsatisfactory.
We have developed an algorithm for accurate and fast triage of ECGs. By transfer learning, feature discovery in diseases where the ECG characteristics are currently unknown, such as primary arrhythmia syndromes and genetic diseases, becomes possible. By improving triage in acute and normal ECGs in the hospital, at home and at the general practitioner, we can decrease unnecessary hospital visits and cardiologist consultations and improve time-to treatment of life-threatening diseases. This will lead to a dual decrease in healthcare burden and costs. In this consortium we aim to develop artificial intelligence (WP1) that will aid clinical care by automatically triaging ECG. Secondly, we will develop a portable multi-lead ECG device that can be used by patients at home and by health care professionals, enabling high quality ECG acquisitions for fast diagnosis. Finally, the developed artificial intelligence will be incorporated in the device to allow instant stand-alone triage and diagnosis of the acquired ECGs.
Our final goal is to validate and implement the developed technology. For WP1 this includes validation studies with a panel of cardiologists and hard outcomes and external validation in a different setting. For WP2 this includes a proof-of-concept study, pivotal study and implementation study leading to CE-certification.
- van de Leur, R. R., Blom, L. J., Gavves, E., Hof, I. E., van der Heijden, J. F., Clappers, N. C., Doevendans, P. A., Hassink, R. J., & van Es, R. (2020). Automatic Triage of 12-Lead ECGs Using Deep Convolutional Neural Networks. Journal of the American Heart Association, 9(10), e015138. https://doi.org/10.1161/JAHA.119.015138
- Van De Leur, R. R., Taha, K., Bos, M. N., Van Der Heijden, J. F., Gupta, D., Cramer, M. J., Hassink, R. J., Van Der Harst, P., Doevendans, P. A., Asselbergs, F. W., & Van Es, R. (2021). Discovering and Visualizing Disease-Specific Electrocardiogram Features Using Deep Learning: Proof-of-Concept in Phospholamban Gene Mutation Carriers. Circulation: Arrhythmia and Electrophysiology. https://doi.org/10.1161/CIRCEP.120.009056
- Bos, M. N., Van De Leur, R. R., Vranken, J. F., Gupta, D. K., Van Der Harst, P., Doevendans, P. A., & Van Es, R. (2020). Automated Comprehensive Interpretation of 12-lead Electrocardiograms Using Pre-trained Exponentially Dilated Causal Convolutional Neural Networks. Computing in Cardiology, 2020-Septe, 2–5. https://doi.org/10.22489/CinC.2020.253
- Vranken, J. F., van de Leur, R. R., Gupta, D. K., Juarez Orozco, L. E., Hassink, R. J., van der Harst, P., Doevendans, P. A., Gulshad, S., & van Es, R. (2021). Uncertainty estimation for deep learning-based automated analysis of 12-lead electrocardiograms. European Heart Journal - Digital Health, 1–41. https://doi.org/10.1093/ehjdh/ztab045
- Rutger R van de Leur, Machteld J Boonstra, Ayoub Bagheri, Rob W Roudijk, Arjan Sammani, Karim Taha, Pieter AFM Doevendans, Pim van der Harst, Peter M van Dam, Rutger J Hassink, René van Es, Folkert W Asselbergs, Big Data and Artificial Intelligence: Opportunities and Threats in Electrophysiology. Arrhythmia Electrophysiol Rev. 2020;9(3):146–54. https://doi.org/10.15420/aer.2020.26