Faraz Hassan
Pronouns: He/Him
Research Mentor(s): Porag Das
Research Mentor School/College/Department: University of Michigan Medical School / Medicine
Program: UROPF
Session: Session 2 (10:00am – 10:50am)
Authors: Faraz Hassan, Anna Shinohara, Yashmeet Kaur, Omar Sohail, Dr. Karandeep Singh, Jeet Das
Abstract
Introduction Cardiac auscultation is an important process of listening to the heart and plays a critical role in the diagnosis of cardiac diseases. Previous studies have shown that, even amongst cardiologists, auscultation alone is only up to 60% accurate in diagnosing cardiac problems. Electronic stethoscopes, on the other hand, provide a better medium to listen to the heart due to possibilities for post-recording analysis and more sensitive microphones that can pick up specific heart sounds. Thus when compared to the human ear, these sophisticated tools provide a much better way to listen to the heart and analyze heart sounds for research. The purpose of this study is to analyze the effect of left ventricular ejection fraction (LVEF) of patients on the audio quality of recorded heart sounds. If a relationship is found, the next step is to build a machine learning algorithm that can predict LVEF based on the heart sound from auscultation at much earlier stages in diagnosis of a patient’s heart. Proper utilization of this algorithm and further understanding of this relationship can be crucial for enhancing patients’ care, diagnosis and treatment in several ways. Methods Hospitalized adult patients at the University of Michigan who have or will undergo echocardiography within ninety days were identified. Heart sounds were collected with a EKO Duo electronic stethoscope at each of the four cardiac valvular positions (aortic, mitral, tricuspid, pulmonic) for 30 seconds each. These heart sounds are analyzed and given a distinct score ranging from 0 or 1 by three different research assistants. This score is given based on whether an S1, S2, or murmurs are heard at the aortic and mitral auscultation points. Lastly these scores are compiled to sum an overall sound quality score for each patient. These scores are then compared against the LVEF of each patient from their echocardiography reports. Results Heart sounds from 200 patients have been randomly selected from the overall pool of 670+ patients with heart recordings. The PCG sounds from these randomly selected patients are being analyzed and scored for sound quality. Initial analysis shows results that have a correlation between heart sound quality and LVEF of patients who have gone through an ECHO over the last 90 days from the time of recording. Conclusion Cardiac auscultation is the initial diagnostic tool for cardiovascular disease. Knowing that even the professionally trained ear cannot diagnose heart problems with higher than 60% accuracy, it is important to determine if machine learning could improve the accuracy of diagnosis. These findings can help better define the relationship between heart sounds and ejection fraction, which may have a clinical impact because patients diagnosed with conditions such as aortic stenosis or heart failure are often observed to have lower LVEF. Hence, earlier recognition of LVEF from heart sounds can indicate these conditions and can alleviate negative clinical outcomes from late recognition of these diseases.
