“Echonet-Dynamic is a great resource for examining echocardiography, one of the core imaging modalities of cardiovascular medicine. The dataset comes with labels for left ventricular tracings and ejection fraction as well as source code for learning about implementing deep learning models and notebooks for converting the images to video format. Furthermore, there are a few other “flavors” of echonet to evaluate other predictions, however, some key weights and disease labels are missing in these.
A moderate amount of legwork is required to get the models running if you are planning to run them on large numbers of echocardiograms. This includes setting up gpu resources (there is an option to run using cpu, but this is not recommended because the run times become significantly longer) and editing some of the source code to work locally as well as implementing methods for selecting appropriate views from echocardiograms (not included with dataset).”
— Tom Liu, Bluhm Cardiovascular Institute Artificial Intelligence Research Fellow in Cardiovascular Disease
