Automated Detection of Reduced Ejection Fraction Using an ECG-Enabled Digital Stethoscope: A Large Cohort Validation

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Asymptomatic left ventricular systolic dysfunction (ALVSD) impacts millions globally, often diagnosed late, leading to poor outcomes.
• Current ALVSD detection methods are insufficient, highlighting the need for advanced diagnostic tools.
• Electrocardiogram-based algorithms are emerging as promising methods for ALVSD detection.

Purpose of the Study:

• To develop and validate a convolutional neural network (CNN) model for detecting low ejection fraction (EF).
• To assess the CNN model's utility using single-lead electrocardiogram and phonocardiogram data from a digital stethoscope.
• To evaluate the model's performance in a large patient cohort.

Main Methods:

• A multicenter observational study enrolled 2,960 adults undergoing echocardiography.
• Digital stethoscopes captured patient data, with echocardiograms performed within one week.
• The CNN model's performance was benchmarked against echocardiographic EF measurements, categorizing EF as normal/mildly reduced (>40%) or moderate/severely reduced (≤40%).

Main Results:

• The CNN model achieved an area under the receiver operating characteristic curve of 0.85.
• Key performance metrics included 77.5% sensitivity, 78.3% specificity, 20.3% positive predictive value, and 98.0% negative predictive value.
• Subgroup analyses confirmed consistent performance across diverse demographics and comorbidities.

Conclusions:

• The CNN model, using a digital stethoscope, provides a noninvasive and scalable approach for early detection of reduced EF (≤40%).
• This AI-driven technology can potentially enable earlier diagnosis and treatment of heart failure.
• Early intervention facilitated by this tool may lead to improved patient outcomes.