Detection of Heart Conditions With Single Lead ECG Using Artificial Intelligence

Heart failure with reduced left ventricular ejection fraction (EF) is a relatively common cardiac pathology with major clinical implications. People with reduced left ventricular EF are at increased risk for sudden death, ventricular and atrial arrhythmias, and acute hemodynamic decompensation due to heart failure. There are proven medical interventions that prevent sudden cardiac death and complications in people with decreased left ventricular EF. Unfortunately, some people with decreased left ventricular EF are asymptomatic, or have non-specific symptoms like dyspnea, and would not receive those interventions in a timely manner. Currently, there are no effective ways to screen for asymptomatic decreased left ventricular EF in the population, because detection of low EF requires the use of echocardiography. There is a significant need to identify novel technologies that can help to detect people with decreased left ventricular EF in a simple, effective, and reliable manner. Eko Devices features a cloud-based platform of point-of-care cardiac screening devices and machine learning algorithms that enables more effective detection and management of cardiovascular disease. In this study, we will use the Eko DUO device to collect single-lead ECG data. The Eko DUO is an FDA-cleared and CE-marked electronic stethoscope that allows audio recording of heart sound to produce a phonocardiogram (PCG) as well as recording a single-lead electrocardiogram (ECG). The DUO features 60x audio amplification, ambient noise reduction, a 4000Hz sample rate, and 4 audio filters. The ECG component is made up of 2 stainless steel electrodes, 0.01Hz high-pass filter, selectable 50/60Hz mains filter, and a 500Hz sample rate. The de-identified auscultatory DUO recordings transmit wirelessly via Bluetooth to the secure, HIPAA-compliant Eko application on a smartphone or tablet, which allows the user to playback heart sound recordings, annotate notes on recorded audio, and save recordings. This data is synced in real-time to a secure, HIPAA-compliant, cloud-based Amazon Web Services (AWS) database server managed by Eko Devices. It has been previously demonstrated that artificial intelligence processing information from a 12-lead ECG can help to identify people with decreased left ventricular EF1. Using paired 12-lead ECG and echocardiogram data, including the left ventricular ejection fraction, from 44,959 patients at the Mayo Clinic, a convolutional neural was trained to identify patients with low ejection fraction. When tested on an independent set of 52,870 patients, the model showed an Area Under the Curve ("AUC") of 0.93 and an accuracy of 86%. We have also developed a single-lead version of the same algorithm, which will be more easily accessible in a clinical setting since it can be used with a single-lead ECG device like the Eko DUO device. We propose to validate performance of this new model using the current study.