Test Characteristics of Smart Phone-Enabled Phonoscopy with Artificial Intelligence-Enhanced Detection of Cardiovascular Disease

Abstract Purpose: Valvular heart disease (VHD) is a major contributor to morbidity and mortality. Auscultation has been a cornerstone of initial VHD diagnosis but requires diagnostic skill to be valuable and is often not performed at all. We therefore sought to explore the test characteristics of an AI-based software, Stethophone AI, that records, augments and visualizes bioacoustic sounds via smartphone microphone in detecting structural and rhythmic cardiac anomalies among a cohort of patients with known VHD. Methods: A random sample of patients presenting to Structural Heart Clinic with known VHD were screened with Stethophone AI for murmur detection as part of routine clinical care, typically using a single point of auscultation Results: Among 68 patients tested, the sensitivity for murmur detection was 95.2%. The positive predictive valve was100% and false negative rate was 4.8% for murmur detection. While the positive predictive value remained consistent across murmur types, sensitivity varied according to valve pathology. The system also demonstrated 100% sensitivity for patients in atrial fibrillation (n = 6). Conclusions: An AI-based system, deployed as a smartphone application that utilizes the device’s built in microphone can detect murmurs of valvular heart disease and atrial fibrillation with excellent discrimination