Physicians have long used separate modalities such as electrocardiograms (ECGs) and magnetic resonance images (MRIs) to diagnose heart conditions. But now, a team of researchers from the Eric and Wendy Schmidt Center at the Broad Institute of MIT and Harvard has developed a machine learning model that can simultaneously analyze both ECGs and MRIs to predict patient characteristics related to the heart, such as its weight and structure, as well as detect heart diseases.
The model developed by the team is an autoencoder, an advanced machine learning algorithm. It takes tens of thousands of ECG and MRI recordings from UK Biobank participants and uses them to create concise representations. This integrated approach allows the model to detect patterns that traditional methods that work on individual data might have missed. The team believes their tool could eventually be used to predict heart conditions with only a routine ECG recording.
The researchers also used the autoencoder representation to generate MRI movies based on just an individual’s ECG recording. Moreover, they used their model to find new genetic markers of heart disease. Autoencoders don’t require data labeled by humans, so the team was able to generate representations that reflected the state of the person’s heart and used the data to look for genetic variants impacting the same state. Their model produced a list of variants, some of which were previously known and some new, for further investigation.
The Broad Institute is an initiative fostering collaboration between the institutes of MIT and Harvard to advance biomedical research. Anthony Philippakis, a senior co-author on the study, is the Chief Data Officer at the Broad, and Caroline Uhler, a co-senior author, is a professor in the Department of Electrical Engineering and Computer Science as well as the Institute for Data, Systems, and Society at MIT.
Adityanarayanan Radhakrishnan is a co-first author on the study, an Eric and Wendy Schmidt Center Fellow at the Broad, and a graduate student at MIT in Uhler’s lab, while Sam Friedman—a senior machine learning scientist in the Data Sciences Platform of the Broad—is the other co-first author. Both have acknowledged the potential of the model in addressing various diseases, and are currently working on applying this autoencoder framework to study neurological diseases.
