EchoGo Heart Failure (2.0)

EchoGo Heart Failure (2.0) is an automated machine learning-based decision support system designed as a diagnostic aid for clinicians assessing adult patients for Heart Failure with Preserved Ejection Fraction (HFpEF). The device utilizes a convolutional neural network to analyze apical 4-chamber echocardiogram views (with ejection fraction ≥50%), providing a binary classification of HFpEF presence and an "EchoGo Score" to support clinical interpretation.

EchoGo Heart Failure (2.0) is a prescription software device that functions as an automated machine learning-based decision support system, designed to aid clinicians in diagnosing Heart Failure with Preserved Ejection Fraction (HFpEF) in adult patients (over 25 years old). It analyzes DICOM files of apical 4-chamber echocardiogram views where the patient has an ejection fraction of ≥50%. The system outputs a binary classification indicating the likelihood of HFpEF and an "EchoGo Score" (0-100%), presented with a comparative histogram against a known population, to support the clinician's assessment; however, it is not intended for sole diagnostic use. The device's core is an artificial intelligence model, specifically a convolutional neural network, which was trained on an enlarged dataset with enhanced pre-processing and data augmentation techniques. Clinical validation was performed via a retrospective case-control study involving 1,578 patients from multiple US sites, showing a sensitivity of 90.3% and specificity of 86.1% (excluding "no classification" studies), with an AUROC of 0.947. The system demonstrated 100% repeatability and high reproducibility (82.6% positive, 82.4% negative agreement), though 7.4% of analyses resulted in "No Classification."

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 EchoGo Score distribution in the provided histogram (compared against the Independent Testing Dataset) influence clinical decision-making beyond the binary classification, particularly for scores near the 50% threshold?; What specific image quality parameters or echocardiographic view variations most significantly contribute to the 7.4% "No Classification" output, and are there actionable guidelines for sonographers to optimize image acquisition for this device?; Considering the AI model was trained on more data with additional pre-processing, what was the specific incremental improvement in sensitivity, specificity, and AUROC attributable to these changes compared to the prior version's performance?; How does the device's diagnostic accuracy (sensitivity, specificity, EchoGo Score calibration) perform across diverse patient subgroups, such as those with varying severities of diastolic dysfunction, different comorbidities (e.g., hypertension, diabetes), or specific demographic factors not explicitly detailed?; What is the recommended clinical pathway or further diagnostic steps when EchoGo Heart Failure 2.0 provides a "No Classification" output, and how does this affect the overall diagnostic workflow for suspected HFpEF?