Smart stethoscope for cardiac auscultation in general practice: a prospective feasibility study of AI-assisted detection of atrial fibrillation, heart failure, and valvular heart disease

ABSTRACT Objectives Artificial intelligence (AI)–enabled digital stethoscopes combine phonocardiography and electrocardiography to support detection of cardiac rhythm and structural abnormalities. This study evaluated the feasibility and exploratory diagnostic performance of AI-guided cardiac auscultation during routine general practice consultations and home visits. Methods In this prospective feasibility study, 50 consecutive patients aged ≥65 years underwent AI-assisted auscultation using the Eko CORE 500 during routine care. Recordings were attempted at four standard cardiac positions. Feasibility outcomes included technical failure, workflow disruption, and proportion of analyzable recordings (defined as successful AI output based on combined ECG and phonocardiography signals). Exploratory diagnostic performance was assessed against previously established diagnoses of atrial fibrillation (AF), heart failure (HF), or valvular heart disease (VHD) documented in the electronic medical record. Results AI-guided cardiac auscultation was completed in all patients without device malfunction or meaningful workflow disruption (median acquisition time 1–2 minutes). At least one analyzable recording was obtained in 47/50 patients (94%), and complete four-position analyses in 42/50 (84%). Signal limitations were mainly attributable to obesity, chest hair, or excess breast tissue. Among 47 analyzable patients, 11 had known AF, HF, or VHD. Sensitivity for detecting these conditions was 81.8% and specificity 91.7%. One new case of clinically relevant mitral regurgitation was identified. Conclusions AI-enabled digital auscultation was feasible in routine general practice, with high rates of analyzable recordings and minimal workflow impact. Larger studies with contemporaneous reference standards are warranted to determine clinical utility.