AI stethoscope doubles detection of serious valve disease in primary care study

By Hugo Francisco de SouzaReviewed by Susha Cheriyedath, M.Sc.Feb 8 2026

New evidence suggests AI-assisted auscultation may help clinicians detect hidden valvular heart disease earlier, potentially reshaping frontline cardiac screening while raising important questions about implementation and diagnostic balance.

Study: Artificial-intelligence-enabled digital stethoscope improves point-of-care screening for moderate-to-severe valvular heart disease. Image Credit: Natali _ Mis / Shutterstock

In a recent prospective study published in the European Heart Journal Digital Health, researchers compared the diagnostic accuracy of primary care providers using standard stethoscopes with that of a relatively novel artificial intelligence (AI) enabled digital stethoscope. The study aimed to determine whether the latter could improve the accuracy of current diagnoses of valvular heart disease (VHD).

The study found that the AI system demonstrated a sensitivity of 92.3% for detecting audible VHD, compared with 46.2% for standard care (P = 0.01). Although the AI tool showed slightly lower specificity, it identified twice as many cases of previously undiagnosed moderate-to-severe disease, suggesting a role as a screening adjunct rather than a replacement for clinical assessment.

Background

Valvular heart disease is a serious cardiac condition in which one or more heart valves, including the aortic, mitral, tricuspid, or pulmonary valves, fail to open or close properly, disrupting blood flow.

Common symptoms include shortness of breath, fatigue, chest pain, and palpitations. Disease prevalence increases with age and is estimated to affect more than half of adults over 65 to some degree, although moderate-to-severe disease is substantially less common.

Diagnosis remains challenging because more than half of patients with clinically significant disease are asymptomatic.

Traditionally, diagnosis relies on clinician-performed auscultation. However, prior research suggests that even experienced general practitioners achieve limited sensitivity when screening asymptomatic patients, contributing to delayed diagnosis and disease progression.

Study Design and Methods

The study explored whether deep learning algorithms, combined with digital acoustic recordings, could help detect cardiac abnormalities that may be missed during routine examinations.

This was a prospective single-arm diagnostic accuracy study conducted across three primary care clinics between June 2021 and May 2023. The cohort included 357 patients aged 50 years and older who were at elevated cardiovascular risk but had no prior diagnosis of VHD or known cardiac murmur.

Risk factors included hypertension, body mass index (BMI) of 30 or higher, diabetes, hyperlipidaemia, atrial fibrillation, prior myocardial infarction, stroke or transient ischemic attack, coronary revascularisation, or other established cardiovascular disease.

Participants underwent two independent screening protocols.

In standard-of-care (SOC) screening, primary care providers (PCP) performed four-point cardiac auscultation using conventional stethoscopes.

In AI-augmented screening, study coordinators recorded phonocardiogram (PCG) data using a digital stethoscope. Recordings were analysed by an AI algorithm cleared by the FDA to detect heart murmurs.

All participants underwent echocardiography to confirm structural heart disease. An independent expert panel reviewed the digital audio recordings to verify the presence of audible murmurs and was blinded to AI results.

Audible VHD was defined as moderate-to-severe disease confirmed on echocardiography, together with an expert-confirmed audible murmur, recognising that some structurally significant disease may not produce a clearly audible murmur.

Study Findings

The AI-augmented system substantially outperformed standard auscultation when detecting audible VHD. Sensitivity was 92.3% for AI compared with 46.2% for SOC screening (P = 0.01).

Among confirmed cases, standard examination missed seven of thirteen patients, whereas the AI system missed only one. For previously undiagnosed moderate-to-severe VHD, the AI identified 12 cases, compared with 6 detected by PCPs.

This increased sensitivity was accompanied by reduced specificity. The AI system demonstrated a specificity of 86.9 percent, compared with 95.6 percent for clinicians (P < 0.001), resulting in more false-positive findings.

Using echocardiography alone as the reference standard for moderate-to-severe disease, regardless of murmur audibility, the AI system still outperformed standard care, with a sensitivity of 39.7 percent versus 13.8 percent for clinicians (P = 0.01).

Conclusions

This study suggests that integrating AI-enabled digital stethoscopes into primary care may substantially improve the detection of VHD compared with traditional auscultation. These tools may function as a second layer of screening support, enabling earlier identification and referral.

Earlier detection does not automatically translate into improved clinical outcomes, as this study evaluated diagnostic accuracy rather than downstream management or prognosis.

Several authors reported affiliations with the device manufacturer, which should be considered when interpreting the findings despite disclosed conflicts of interest.

Lower specificity may increase echocardiography referrals and healthcare utilisation, underscoring the need for future cost-effectiveness analyses.

Limitations include a modest sample size, limited geographic scope, incomplete demographic detail, and lack of systematic symptom assessment. Despite these constraints, the findings indicate that AI augmentation may represent a meaningful advance in point-of-care cardiac screening.