Development of an Algorithm to Differentiate Heart Murmurs Using Electronic Stethoscopes

Valvular heart disease (VHD), a subclass of structural heart disease (SHD), is a major cause of mortality and reduced quality of life for tens of millions of patients worldwide. Detecting VHD early allows the disease to be carefully monitored so that surgery can be considered as soon as the stenosis progresses to severe. Unfortunately, it is difficult for many clinicians to accurately identify VHD in its early stages and assess its severity through physical examination alone. This difficulty stems from both the subtlety of symptoms associated with VHD, and from clinicians' inability to reliably detect the heart murmurs indicative of VHD using a traditional stethoscope, particularly in the noisy and rushed environments characteristic of many healthcare settings.

AS, mitral regurgitation (MR), and tricuspid regurgitation (TR) all result in reliably reproducible auscultatory findings. Yet despite this, auscultation as a diagnostic procedure is notoriously poor: its accuracy to detect AS, MR, and TR ranges only from 5-40%. In addition, diastolic and continuous murmurs, which are rarer than the systolic murmurs caused by AS, MR and TR, and almost always pathologic, are usually very difficult to detect by frontline clinicians. Diastolic murmurs are typically caused by VHD such as aortic regurgitation (AR), pulmonic regurgitation (PR), mitral stenosis (MS), or tricuspid stenosis (TS), and continuous murmurs are typically caused by non-valvular forms of SHD, such as congenital aortopulmonary window, shunts, and patent ductus arteriosus (PDA). Missing one of these murmurs could lead to significantly delayed diagnosis and care.

Because of how difficult it is for clinicians to reliably screen for SHD through history taking and physical examination, the diagnosis of SHD relies heavily on echocardiography (echos), which is accepted as the gold standard for diagnosis of SHD severity by the cardiology community. The consequence of this is several-fold. First, many patients receive echos that would not be necessary if better screening and assessment tools existed. Second, echocardiography makes up a disproportionately large segment of healthcare expenditure. Third, patients are burdened with the responsibility of getting repeated echos due to the lack of better screening tools. The emotional, physical, and financial burden that accompanies such excessive testing, especially for patients living in underserved communities, cannot be overstated.

Overall, there is a growing need for better preliminary screening tools that would identify these conditions, even when asymptomatic, early on and more consistently, to a degree of accuracy comparable to gold standard echocardiography. Such a tool would improve clinicians' confidence in their assessment during physical exams, which would enable them to help their patients make informed care decisions, offer disease-altering therapies, and at the same time reduce the reliance on, and frequency of, unnecessary and expensive auxiliary workups. Lastly, by facilitating early detection of VHD, morbidity and mortality rates would be expected to decrease due to earlier intervention.

The FDA-approved Eko CORE and Eko DUO electronic stethoscopes offer clinicians a familiar and inexpensive tool that is widely accepted by patients, while at the same time offer sensors and technology that can improve screening and detection of medical conditions such as valve disease, as compared to auscultation with the traditional stethoscope. Both CORE and DUO feature sound amplification during auscultation, which improves the ability for the clinician to detect nuanced changes in heart sounds and patterns. The CORE allows recording of the audio to produce a phonocardiogram (PCG), while the DUO is able to record audio to produce a PCG as well as simultaneous recording of a single-lead electrocardiogram (ECG). Both devices use Bluetooth to wirelessly transmit PCG and ECG data to the Eko mobile application, which allows clinicians to visualize PCG and ECG as they auscultate, as well as to playback heart sound recordings, annotate notes on recorded audio, and save and share recordings. These features uniquely situate Eko electronic stethoscopes to be highly effective as preliminary screening tools in clinical practice.

Eko Devices has developed and received 510(k) FDA approval for a machine-learning (ML) algorithm that can detect the presence or absence of heart murmur in adult patients with VHD, using either the Eko CORE or DUO electronic stethoscopes to record heart sounds. This algorithm was built using a deep neural network model trained on 5,878 PCG recordings collected on Eko CORE and Eko DUO from over 5,318 unique patients to classify heart sound recordings as containing a heart murmur or containing no audible murmur, and to discriminate good quality heart sound recordings from poor quality heart sound recordings. The trained model has been independently validated in a multi-site clinical study that enrolled 681 VHD patients with matched gold-standard echocardiograms. FDA approval was received in January 2020.

The present study seeks to expand murmur detection to include VHD classification through the development of novel ML algorithms that are able to distinguish between systolic vs. diastolic vs. continuous murmurs, as well as classify VHD type and severity, using 4-point auscultation with Eko CORE and DUO electronic stethoscopes to record heart sounds. These new algorithms will integrate the time-synchronized ECG streams from DUO to improve distinction between S1 and S2 heart sounds and improve performance of identifying systolic vs. diastolic time intervals. When combined with the FDA-approved algorithm for signal quality and murmur detection, these new algorithms will enable clinicians to have a complete VHD decision support system, as based on ACC/AHA Patient Management Guidelines, when using Eko stethoscopes in their clinical practice.20 This will arm clinicians with information to help them detect SHD early, make clinical care decisions with confidence and consistency, and improve overall patient outcomes while decreasing cost and increasing efficiency.