Artificial intelligence (AI)-based stethoscopes can accurately diagnose left-sided valvular heart disease (VHD), according to original research published in JACC: Advances.

"Reliance on echocardiography over the past decades has led to underuse of auscultation by the clinician. It should not be ignored that cardiac auscultation plays an extremely important auxiliary role in the diagnosis of VHD," write study authors Ziwei Zhou, MD, et al., who used the smartho-D2 electronic stethoscope to record heart sounds from 514 patients in a quiet clinical environment. They used the recordings of 304 patients (mean age 64 years; 39% women) to train a machine learning algorithm that was then tested against echocardiography for the 210 additional participants (mean age 62 years; 41% women).

Results within the training phase showed that the algorithm could accurately identify patients with left-sided VHD (area under the receiver operating characteristic [AU-ROC], 0.85), a composite of left-sided aortic stenosis (AU-ROC, 0.76), aortic regurgitation (AU-ROC, 0.71), mitral stenosis (AU-ROC, 0.64) and mitral regurgitation (AU-ROC, 0.79).

Sensitivity and specificity of the VHD diagnostic results in the testing phase were 70% and 74%, respectively – with an AU-ROC of 0.76.

"It is crucial to early recognize VHD," write Zhou and colleagues, stating that it "is becoming more prevalent with increasing life expectancy and plays as an important driver of cardiovascular morbidity and mortality."

They add that, compared to the longer process of echocardiography, this method takes approximately 91 seconds, indicating that "our models may make routine screening for VHD more practical and promote the early identification and diagnosis of patients with left-side VHD."