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

Using a custom-made web platform, expert annotators listened to heart sound recordings with headphones while viewing a plot of the PCG. Expert annotators were cardiologists having completed fellowship training in cardiology and each with at least 10 years of clinical practice. Two sensory modalities helped annotators assess signal quality (1-5 scale with defined rubric) and systolic and diastolic murmur presence (true or false for each) and grade (Levine scales 1-6 or 1-4 22 ). Since murmur grade was determined by recording only, systolic murmur grades 5 and 6 were not used. To establish a single set of ground truth labels for a recording we aggregated the responses of the three cardiologists. For a binary presence of each murmur type, we used a majority vote. For signal quality and murmur grade, both of which were optional, we used the median of the responses if there were three and the lessor if there were only two.