Abstract: Heart failure (HF) is diagnosed using echocardiographic indices of filling pressure and systolic function, yet echocardiography is unavailable in most first-contact settings where patients with dyspnea first present. This paper describes a dual-signal smart stethoscope that simultaneously records chest phonocardiography and finger photoplethysmography. The stethoscope system is driven by a multimodal deep-learning model that estimates HF-relevant echocardiographic indices—pulmonary artery systolic pressure (PASP), E/e′, left atrial volume index (LAVI), and left ventricular ejection fraction (EF)—from a single 20-second recording. The model was trained on 1,828 recordings from 1,635 patients paired with same-day echocardiography at a tertiary hospital and evaluated on an independent held-out cohort of 118 recordings from 96 patients. The estimated indices discriminated guideline cut-offs with areas under the ROC curve of 0.81 for PASP ≥ 40 mmHg, 0.79 for LAVI > 34 mL/m², 0.77 for EF ≤ 40%, 0.78 for EF < 50%, and 0.66 for E/e′ > 14. Two pre-registered screening axes—a filling-pressure axis combining PASP, E/e′, and LAVI, and an EF axis—showed a double dissociation: the pressure axis identified HF with preserved EF physiology (AUROC 0.75 vs 0.37 for the EF axis), whereas the EF axis identified reduced EF (0.77 vs 0.62). A lightweight model deployed on the smartphone reproduced these results (correlation 0.98–0.99 with the research ensemble), and the two axes are presented in the clinical application as a staged display (normal, pre-HF, HF suspected). A 20-second bedside recording can therefore flag HF physiology before echocardiography, supporting point-of-care triage and earlier referral.
Keywords: heart failure; phonocardiography; photoplethysmography; echocardiography; multimodal deep learning; smart stethoscope

Biography: Qun Wei is an Associate Professor in the Department of Biomedical Engineering at Keimyung University in Daegu, South Korea. He holds a Ph.D. in electrical and electronics engineering from Kyungpook National University.
His research focuses on wearable biosignal measurement systems and artificial intelligence for cardiovascular health. He develops multimodal approaches that combine heart sounds and pulse waves to assess cardiac function non-invasively, making cardiovascular assessment simple and accessible in primary care and at home.
In addition, Prof. Wei has published more than 20 SCI-indexed papers and holds over 20 Korean patents, as well as a granted U.S. patent.
He is also Co-CEO of Clairaudience Inc., a faculty startup he co-founded with cardiologist Prof. In-cheol Kim. Through collaboration with the Cardiovascular Center at Keimyung University Dongsan Hospital, his team conducts clinical studies comparing patient heart-sound and pulse-wave recordings with echocardiography in valvular heart disease and heart failure.