Eko Foundation Analysis Software with Transformers (EFAST)

EFAST is a cloud-based software that analyzes heart sounds (phonocardiograms) and ECG recordings to detect structural murmurs, determine murmur timing (systolic/diastolic), identify S1 and S2 heart sounds, and detect atrial fibrillation and sinus rhythm. It provides clinical decision support to healthcare professionals and is intended for use on pediatric and adult patients for murmur detection, with atrial fibrillation detection limited to adults over 18 years. The software is not a sole means of diagnosis and is designed to support, not replace, clinician judgment.

EFAST uses deep neural network machine learning algorithms operating on Eko Digital Stethoscope recordings stored in the cloud (AWS S3). It includes five algorithm components: heart sound signal quality classifier, structural murmur classifier, heart sound timing and murmur timing detector, ECG rhythm classifier (with four classifications), and heart rate calculator. Compared to the predicate EMAS, EFAST adds ECG rhythm classification capability and heart rate calculation functionality that were not present in the predicate.

IEC 62304 standard for software unit, integration, and system-level verification testing; cybersecurity testing. Validation included retrospective studies on proprietary datasets from Eko CORE and CORE 500 Digital Stethoscopes with cardiologist annotations and echocardiogram gold standard for murmur classification, plus multi-expert ECG annotations for rhythm classification.

EFAST has the same intended use as predicate EMAS (structural murmur detection and characterization in pediatric and adult patients) and both are classified as stethoscopes under 21 CFR 870.1875. Although EFAST adds ECG rhythm classification and heart rate calculation features beyond EMAS, the applicant argues these differences do not raise new safety or effectiveness questions because: (1) the predicate EAS already possessed ECG analysis capabilities, providing a reference for safe ECG functionality; (2) structural murmur performance is comparable (EFAST sensitivity 83.4% vs EMAS 85.8%, specificity 86.0% vs 84.5%); and (3) atrial fibrillation detection actually exceeds the reference device EAS (EFAST sensitivity 94.7% vs EAS 73.1%). The device underwent the same validation approach and same regulatory classification as its predicates.