Mount Sinai researchers have developed an Artificial Intelligence (AI) algorithm capable of learning how to identify certain changes in electrocardiograms (ECG), and predict heart failure.
A published study in Journal of the American College of Cardiology: Cardiovascular Imaging, showed the results of this AI algorithm, which would be able to learn and identify ECG changes to alert doctors about patients with possible heart failure.
Lead author of this study, Dr. Benjamin S. Glicksberg, assistant professor of genetics and genomic sciences, member of the Hasso Plattner Institute for Digital Health at Mount Sinai, explained the following about this breakthrough: “We show that deep learning algorithms can recognize blood pumping problems on both sides of the heart from ECG waveform data. Diagnosing these types of heart conditions typically requires expensive and time-consuming procedures. We hope that this algorithm will enable faster diagnosis of heart failure."
To conduct the study, the researchers programmed a computer so that it could read ECGs from patients, and other data extracted from written reports containing Echocardiogram (ECHO) results from the same patients. In this way the written reports were taken as a standard set of data for the computer to compare with the ECG data. In this way it was possible to detect weaker hearts that could suffer from heart failure.
Heart failure is a condition that causes the heart to pump less blood than the body requires. Diagnoses are usually made through ECG and ECO to find out if the patient suffers from this condition. However, the main problem with this type of evaluation or diagnosis is that it requires specialized equipment and medical professionals that are not found in any hospital. Hence the importance of this advance through AI.
For the study, the computer analyzed more than 700,000 ECG and ECHO reports and reports from 150,000 patients in the Mount Sinai Health System between 2003 and 2020. The data was obtained from four hospitals as part of the algorithm's learning process. and to test it we used data from a fifth hospital.
The results showed 94% accuracy in which patients had a healthy ejection fraction and 87% accuracy in predicting those who had an ejection fraction below 40% (normal ejection fraction is 50% or more and weak hearts equal to or minus 40%). Thus, results indicated that the algorithm was effective in predicting heart failure.
"Our results suggest that this algorithm could be a useful tool to help clinicians combat heart failure in a variety of patients," added Dr. Glicksberg.
