EchoGo Core

EchoGo Core is a software application that utilizes machine learning algorithms to automatically analyze echocardiograms and quantify key cardiac parameters such as Ejection Fraction, Global Longitudinal Strain, and Left Ventricular volume. Its methodology aims to support physician diagnosis by providing standardized, reproducible measurements of cardiovascular function in adult patients, notably reducing operator-dependent variability in contouring.

EchoGo Core is a standalone software application designed for the automated quantification and reporting of cardiovascular function in adult populations, serving as an aid to physician diagnosis. The system processes anonymized echocardiogram datasets using a machine learning-based algorithm for auto-contouring of the left ventricular (LV) endocardium. These automatically generated contours, following review and approval by trained operators, are utilized to calculate critical cardiac parameters including Ejection Fraction (EF) via Simpson’s Biplane Method (SBM), Global Longitudinal Strain (GLS) averaged from two and four-chamber views, and LV volumes also using SBM, with results compiled into a clinician's report. Performance evaluations demonstrated EchoGo Core's capabilities, particularly in comparison to predicate devices like TomTec Arena TTA2, highlighting improvements in consistency. A significant distinction is EchoGo Core's use of machine learning and its preclusion of manual contour editing, a feature available in the predicate. This design choice, supported by an input image quality control procedure, successfully eliminated inter- and intra-operator variability (0% bias and RMSE). In contrast, the predicate required manual editing in up to 100% of cases, while EchoGo Core necessitated complete contour rejection in up to 29% of instances. A retrospective validation study involving 378 patient studies confirmed that EchoGo Core met its primary endpoints, achieving Root Mean Square Error (RMSE) values of 5.02% for EF, 2.89% for GLS, 8.0 ml for end-diastolic volume, and 11.1 ml for end-systolic volume compared to predicate-derived reference values.

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 machine learning algorithm in EchoGo Core adapt to variations in image quality across different ultrasound scanner platforms, and what is the impact on contour accuracy and the 29% rejection rate?; What specific features or image characteristics trigger a 'poor contour tracking' rejection by EchoGo Core, and are there plans to refine the algorithm to reduce this rejection rate?; Given that EchoGo Core eliminates manual editing and inter/intra-operator variability, how does its diagnostic accuracy for specific cardiac conditions compare to systems allowing manual adjustments by expert clinicians?; What were the 'set thresholds' for Root Mean Square Error (RMSE) in the performance testing, and how do the achieved RMSE values (e.g., 5.02% for EF) clinically correlate with diagnostic confidence and patient outcomes?; How does the review and approval process by 'trained operators' for the automated contours ensure that the machine learning output is clinically sound, especially in complex or atypical cardiac anatomies?