A Deep Learning Algorithm for Automated Cardiac Murmur Detection Via a Digital Stethoscope Platform

John S Chorba; Avi M Shapiro; Le Le; John Maidens; John Prince; Steve Pham; Mia M Kanzawa; Daniel N Barbosa; Brent E White; Jason Paek; Sophie G Fuller; Grant W Stalker; Sara A Bravo; Dina Jean; Subramaniam Venkatraman; Patrick M McCarthy; James D Thomas

The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/2020.04.01.20050518 doi: medRxiv preprint sensitivity of 97.5% and specificity of 77.7%. These numbers are more sensitive than all the expert cardiologist annotators and more specific than all but one of them. The algorithm detects moderate-to-severe or greater mitral regurgitation with a sensitivity of 64.0% and a specificity of 90.5%, which is more specific but less sensitive than the annotators. Notably, as "algorithms" themselves, the annotators fall nearly directly on the ROC curve of the deep learning algorithm, as applied to mitral regurgitation. As seen for both detection of aortic stenosis and mitral regurgitation, there is considerable variability in performance amongst annotators, even amongst highly trained experts. This likely represents variability in real-world practices amongst clinicians, which is a considerable challenge. This emphasizes the need for a real-time tool that could obviate this inter-observer variability, and the algorithm tested here, when paired with an electronic stethoscope, does exactly this.