Norah Saraya
Pronouns: she/her
Research Mentor(s): Karandeep Singh
Co-Presenter:
Research Mentor School/College/Department: Internal Medicine and Learning Health Sciences / Medicine
Presentation Date: April 20
Presentation Type: Poster
Session: Session 2 – 11am – 11:50am
Room: League Ballroom
Authors: Norah Saraya, Nicole Carpentiere, Jeet Das, Jonathon McBride, Karandeep Singh
Presenter: 72
Abstract
Objective: Valvular heart disease (VHD) is a group of conditions where at least one of the four heart valves are not working properly and affects around 8 million Americans. Cardiac auscultation via a stethoscope has been utilized for centuries to identify the presence of VHD, but the accuracy of diagnosis remains inconsistent leading to adverse outcomes for patients. Machine learning is an area of artificial intelligence that makes use of data to mimic the human learning process and has the potential to aid physicians in clinical decisions. We aim to construct a diagnostic tool by using machine learning techniques in order to improve the accuracy of cardiac auscultation. Methods: The heart sounds and echocardiograms of hospitalized adult patients (>18 years old) at Michigan Medicine who have or will receive echocardiograms within 72 hours were recorded utilizing the Eko DUO digital stethoscope. The sounds were auscultated on four spots on the chest (aortic, pulmonary, tricuspid, and mitral valve positions) for 30 seconds each. The sounds were then uploaded from the Eko application onto a secure database, and machine learning will be used in the construction of the diagnostic tool. Results: 509 patients are enrolled thus far and over 3000 total heart sounds were recorded and uploaded into the database. Data collection remains in progress, as we aim to record the sounds of at least 1000 patients. Assessment of patient demographics, quality of heart sounds, and development of a diagnostic tool is ongoing. Conclusion: Cardiac auscultation remains a fundamental technique for the diagnosis of VHD, yet its accuracy is lacking. Currently, we are collecting patient heart sounds that will be used in machine learning to distinguish heart sounds in order to diagnose VHD. We aim to refine the accuracy of auscultation which will ultimately improve patient outcomes.
Biomedical Sciences, Interdisciplinary
