Skip to content

New Paper on Deep Learning Analysis of COVID-19 Patient ECGs

We are extremely excited to share news of our article’s publication in CVDHJ! The article is open access so should be freely available to all!

In this study, we show that neural networks can use intake ECGs from patients hospitalized with confirmed COVID to predict adverse cardiovascular outcomes or death. We emphasize that the predictive power of these models is quite modest, but not dramatically inferior to standard regression models based on demographics, comorbidities, and machine-extracted ECG metrics. We believe performance of this machine learning approach might be enhanced by combining it with conventional metrics; but that analysis was beyond the scope of the current work. The silver lining is that we now have a robust framework for deep learning solutions, implemented on UW’s HYAK supercomputer system, and we’re extremely excited to apply this to other projects.

Thanks to Zih-Hua (Amber) Chen for her extremely hard work on this project and to our key clinical collaborator, Dr. Arun Sridhar of the UW Cardiac EP service.