Publications & Clinical Research
Eko technology and algorithms are featured in many areas of clinical research. Please contact us if you would like to use Eko products in research.
AHA (2019) - Prospective Analysis of Utility of Signals From an ECG-Enabled Stethoscope to Automatically Detect a Low Ejection Fraction Using Neural Network Techniques Trained From the Standard 12-Lead ECG
​ECG-enabled stethoscopes (ECG-steth) can acquire single lead ECGs during cardiac auscultation, and may facilitate real-time screening for pathologies not routinely identified during physical examination (eg, arrhythmias). We previously demonstrated an artificial intelligence (AI) algorithm applied to a 12-lead ECG (ECG-12) can identify low ejection fraction (EF) (defined as <=35%) with an accuracy of 87%. It is unknown if AI algorithms trained from ECG-12 can be applied to single lead ECGs acquired through devices such as ECG-steth.

Z ATTIA, J DUGAN, J MAIDENS, A RIDEOUT, F LOPEZ-JIMENEZ, P A NOSEWORTHY, S ASIRVATHAM, P A PELLIKKA, D J LADEWIG, G SATAM, S PHAM, S VENKATRAMAN, P FRIEDMAN, S KAPA
​AHA (2019) - Handheld Wireless Digital Phonocardiography for Machine Learning-Based Detection of Mitral Regurgitation
​Background: Mitral regurgitation (MR) is a common disease which can be detected as a murmur on auscultation, but studies show that the majority of new primary care physicians do not detect MR murmurs which are confirmed by transthoracic echocardiography (TTE). The FDA-approved Eko CORE device is a digital stethoscope wirelessly paired with the Eko Mobile application to allow recording and analysis of phonocardiograms (PCG). These PCG data drive a machine learning-based detection algorithm to identify clinically significant MR, validated by TTE, as part of the ongoing Phono- and Electrocardiogram Assisted Detection of Valvular Disease (PEA-Valve) Study. Methods: Patients undergoing TTE at Northwestern Medicine underwent PCG recording by the Eko CORE device. Recordings 15 seconds long were obtained at four standard auscultation positions. A TensorFlow-based machine learning algorithm assessed the presence or absence of murmur with dominant localization to the cardiac apex indicating clinically significant MR, defined as moderate or greater on TTE (Figure 1). Results: To date, 234 patients with 626 recordings have been enrolled, with 32 patients (13.7%) found to have significant MR on TTE. The receiver-operating characteristic curve had an area of 0.764, yielding a sensitivity of 61.5% (95% CI, 42.9-80.0%) and a specificity of 86.3% (95% CI, 76.5-94.7%) for the detection of MR (Figure 2). Conclusion: PCG assessment using the Eko CORE device and machine learning interpretation is a fast and effective method to screen for significant MR and should be validated in a primary care setting, which may lead to more appropriate referrals for echo.

B E WHITE, A M SHAPIRO, M M KANZAWA, S VENKATRAMAN, J PAEK, S PHAM, J MAIDENS, J D THOMAS, P M MCCARTHY
​AHA (2019) - Prospective Analysis of Utility of Signals From an ECG-Enabled Stethoscope to Automatically Detect a Low Ejection Fraction Using Neural Network Techniques Trained From the Standard 12-Lead ECG
​There have been over 15,000 continuous flow left ventricular assist devices (LVADs) implanted in the United States1. As the care of these patients expands into the general community, it is important for providers at all levels to be familiar with the sound of a normal LVAD. The sound generated is normally described as an “LVAD hum”. That non-descriptive term may be misunderstood as all LVADs or “hums” are the same, when in fact the sound produced is unique to each device. Another common misconception held by some providers, is the absence of heart sounds in a normally functioning LVAD. Using apex phonocardiography we were able to better visualize these unique characteristics (Figures 1-4, phonocardiograms of three United States Food and Drug Administration approved durable LVADs), and suggest a refined description of each device sound. The recordings were made on normal functioning LVADs using the EKO CORE Stethoscope Attachment (EKO, Berkeley, CA).

F ARAJ, A AMIN, J COX, P MAMMENMAHA
AHA (2019) - Handheld Wireless Digital Phonocardiography for Machine Learning-Based Detection of Aortic Stenosis
​This research was presented by Dr. Bent White at the 2019 American Society of Echocardiography Scientific Sessions, where he received the 2019 ASE Foundation Top Investigator Award. PCG assessment using the Eko CORE Digital Stethoscope and machine learning interpretation is a fast and effective method to screen for significant AS and should be validated in a primary care setting, which may lead to more appropriate referrals for an echocardiogram.

B E WHITE, J PAEK, S PHAM, J MAIDENS, P M MCCARTHY, J D THOMAS
​American Journal of Cardiology (2019) - The Digital Stethoscope – Two Senses Are Better Than One
​We read with great interest the recent article by Silverman and Balk, and agree that the sound quality of the digital stethoscope is just as good, or superior to an analog stethoscope.1 This is important to today’s practice of medicine where cardiovascular physical examinations are abridged, poorly executed, and minimal effort is undertaken to optimize the auscultatory milieu for a higher yield exam.2 Amplification of sound does not necessarily result in clearer appreciation of heart sounds and murmurs as motion artifacts and surrounding noises are amplified as well.

F G ARAJ MD , J COX BSN RN
​Nature Medicine (2019) - Screening for Cardiac Contractile Dysfunction Using an Artificial Intelligence–Enabled Electrocardiogram
Asymptomatic left ventricular dysfunction (ALVD) is present in 3–6% of the general population, is associated with reduced quality of life and longevity, and is treatable when found. In this work, researchers showed that application of artificial intelligence (AI) to the electrocardiogram (ECG) could identify ALVD. Using paired 12-lead ECG and echocardiogram data, from 44,959 patients at the Mayo Clinic, they trained a convolutional neural network to identify patients with ventricular dysfunction, defined as ejection fraction ≤35%, using the ECG data alone. When tested on an independent set of 52,870 patients, the network model yielded values for the area under the curve, sensitivity, specificity, and accuracy of 0.93, 86.3%, 85.7%, and 85.7%, respectively.

Z ATTIA, MS; J DUGAN, CRC; P A FRIEDMAN, MD; F L JIMENEZ, MD; P A NOSEWORTHY, MD; S J ASIRVATHAM, MD; P A PELLIKKA, MD; D J LADEWIG, BS; G A SATAM, MBA; S KAPA MD
​AHA (2019) - Prospective Analysis of Utility of Signals from an ECG-enabled Stethoscope to Automatically Detect a Low Ejection Fraction Using Neural Network Techniques Trained from the Standard 12-lead ECG
​ECG-enabled stethoscopes (ECG-steth) can acquire single lead ECGs during cardiac auscultation, and may facilitate real-time screening for pathologies not routinely identified during physical examination (eg, arrhythmias). We previously demonstrated an artificial intelligence (AI) algorithm applied to a 12-lead ECG (ECG-12) can identify low ejection fraction (EF) (defined as <35%) with an accuracy of 87%. It is unknown if AI algorithms trained from ECG-12 can be applied to single lead ECGs acquired through devices such as ECG-steth. To demonstrate that an AI algorithm trained using ECG-12 can be applied to ECG-steth for detection of low EF.

Z ATTIA, MS; J DUGAN, CRC; P A FRIEDMAN, MD; F L JIMENEZ, MD; P A NOSEWORTHY, MD; S J ASIRVATHAM, MD; P A PELLIKKA, MD; D J LADEWIG, BS; G A SATAM, MBA; S KAPA, MD
​Journal of Cardiac Failure (2018) - Use of Ventricular Assist Device Acoustical Signatures to Detect Device Thrombosis

Pump thrombosis (PT), the most common cause of failure in ventricular assist devices (VADs), can be mitigated with earlier diagnosis before triggering any clinical events or requiring VAD replacement. Applying machine learning algorithms on pump acoustical signals provides substantial information regarding thrombosis states, and potentially allows for sensitive detection of PT assisting hemolysis biomarkers and pump powers.
B SEMIZ, S HERSEK, M B POUYAN, C PARTIDA, L BLAZQUEZ, V SELBY, G WIESELTHALER, O T INAN, L KLEIN
​Pediatric Cardiology (2019) - Real-World Evaluation of the Eko Electronic Teleauscultation System
Numerous Children's Hospital cardiologists improved murmur detection using Eko's audio amplification and visualization. Their study demonstrates that Eko recordings had a high percent of agreement with in-person auscultation findings and echocardiogram findings, with moderate inter-rater reliability. It was useful in identifying patients with pathologic murmurs who would benefit from further assessment. It was able to discern major types of pathological murmurs. Certain qualitative differences in the recorded sounds as compared to in-person auscultation were identified by the reading cardiologists. They were able to acclimate to these subtle differences. The system was felt to be easy to use, and most cardiologists in the study would consider using it in clinical settings.

S BEHERE, J M BAFFA, S PENFIL, N SLAMON
​Circulation (2018) - Artificial Intelligence Detects Pediatric Heart Murmurs with Cardiologist-Level Accuracy
​This validation study on Eko’s algorithm used a deep neural network to flag pediatric heart murmurs in a noisy hospital setting. A retrospective analysis of audio recordings from 54 patients collected at duPont Hospital for Children using an Eko Core digital stethoscope was performed. This murmur detection algorithm outperformed 4 out of 5 pediatric cardiologists at murmur detection.

N SLAMON MD, J MAIDENS PHDIEEE
IEEE (2018) - Deep Unsupervised Representation Learning for Abnormal Heart Sound Classification
​This study presents a comparison of conventional and state-of-the-art deep learning based computer algorithm paradigms for the audio classification task of normal, mild abnormalities, and moderate/severe abnormalities as present in phonocardiogram recordings. In particular, they explore the suitability of deep feature representations as learnt by sequence to sequence autoencoders based on the auDeep toolkit.

S AMIRIPARIAN, M SCHMITT, N CUMMINS, K QIAN, F DONG, B SCHULLER