Eko Murmur Analysis Software (EMAS)

The Eko Murmur Analysis Software (EMAS) is a cloud-based system utilizing machine learning and signal processing to analyze heart sounds from digital stethoscopes, providing clinicians with decision support for identifying and classifying heart murmurs. Its methodology involves API-driven data submission, algorithmic detection of S1/S2 sounds, murmur presence, character (innocent vs. structural), and timing, aiming to enhance the evaluation of heart health in pediatric and adult patients.

The Eko Murmur Analysis Software (EMAS) is a cloud-based clinical decision support tool designed to assist clinicians in evaluating patient heart sounds. It analyzes phonocardiogram (PCG) and optional electrocardiogram (ECG) data, acquired via Eko DUO or Eko CORE digital stethoscopes and transmitted through an application programming interface (API). The software employs signal processing techniques and machine learning algorithms to assess PCG signal quality, detect the presence and timing of S1 and S2 heart sounds, and identify systole and diastole regions. A key function of EMAS is its ability to detect heart murmurs, classify them as either innocent (non-pathologic) or structural (pathologic), and determine their timing within the cardiac cycle. The system outputs results in a JSON format for display on a user's application, intended for use by clinicians on both pediatric and adult patients as an adjunct to their clinical judgment, not as a standalone diagnostic tool. The performance of EMAS was validated through a retrospective clinical study on a proprietary database of 2,380 heart sound recordings from 615 unique subjects (age 0-95 years), collected using Eko CORE and Eko DUO digital stethoscopes. Ground truth was established via cardiologist annotations paired with echocardiograms. In murmur classification, EMAS achieved 85.6% sensitivity (95% CI: 82.6-88.7) and 84.4% specificity (95% CI: 81.3-87.5); the clinical validation demonstrated that the lower bounds of these 95% confidence intervals were above 75.0%, indicating its effectiveness relative to established benchmarks. For heart sound timing, S1 detection showed 96.2% sensitivity and 97.1% PPV, while S2 detection achieved 92.3% sensitivity and 94.3% PPV. No adverse events occurred during the study, supporting the software's capability for effective heart sound analysis.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the algorithm's performance in murmur classification vary across different age groups, particularly between neonatal, pediatric, and adult populations, given the diverse age range (0-95 years) in the validation study?; What specific machine learning models and features derived from the PCG/ECG signals are most critical for the accurate distinction between innocent and structural murmurs?; Given the predominantly white (98% of the 45% where race was recorded) validation cohort, what steps are planned to validate or adapt the EMAS algorithms for more racially diverse populations to ensure equitable performance?; How does the Signal Quality Detection Algorithm handle common sources of interference in clinical settings (e.g., patient movement, ambient noise, poor stethoscope placement), and what is the impact of 'poor' signal quality classification on subsequent murmur analysis?; What is the typical processing time for a recording from data upload via the API to the return of the JSON analysis results, and how might this vary with ECG data inclusion or signal complexity?