EchoGo Pro

EchoGo Pro is a machine learning-based software designed to assist physicians in assessing coronary artery disease by automatically analyzing stress echocardiogram images. It utilizes DICOM datasets to perform left ventricular segmentation and calculate geometric parameters, which are then processed by a pre-trained classification model to provide a categorical assessment of significant CAD likelihood.

EchoGo Pro v1.0.2 is a standalone software application designed as an adjunct to diagnostic stress echocardiography for assessing coronary artery disease (CAD). It processes anonymized DICOM 3.0 compliant stress echo (SE) datasets, utilizing standard clinical protocols with apical 2-chamber, 4-chamber, and parasternal short-axis views at rest and peak stress. The software automatically registers images, segments the left ventricle (LV) by detecting endocardial contours, and allows trained operators to review and approve these contours. Approved contours are then used to calculate geometric parameters, which are fed into a fixed, pre-trained machine learning classification model to assess the possibility of significant CAD, defined as ≥70% stenosis in major coronary arteries. Performance testing demonstrated that EchoGo Pro improved diagnostic accuracy and inter-operator agreement. When readers used EchoGo Pro, the area under the receiver operating characteristic curve (AUROC) showed a mean improvement of 0.054 (95% CI 0.032, 0.077, p=0.02) compared to unaided interpretation, exceeding the mean reader improvement reported for the predicate device (0.034). The native system performance of EchoGo Pro achieved an AUROC of 0.927, with a specificity of 0.927 (95% CI 0.878, 0.976) and a sensitivity of 0.844 (95% CI 0.739, 0.950), which was greater than the native system performance reported for the predicate device (0.882). Furthermore, inter-clinician agreement, measured by the Kendall Tau-B correlation coefficient, significantly increased from an average of 0.58 (unaided) to 0.82 (aided by EchoGo Pro) (p<0.001), also exceeding the improvement reported for the predicate device (0.1393).

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. What specific geometric parameters calculated from the LV contours are most heavily weighted by the classification model in determining CAD likelihood?; How was the 'known outcome' dataset for training the classification model curated, and what measures were taken to ensure its diversity and representativeness of the intended use population, especially considering the 6-month window for invasive angiography?; What is the impact of suboptimal image quality or variability in input DICOM stress echo datasets (e.g., from different ultrasound machines or sonographer skill levels) on the performance of the auto-contouring algorithm and the final CAD assessment?; Beyond the reported AUROC, sensitivity, and specificity, what are the positive predictive value (PPV) and negative predictive value (NPV) of EchoGo Pro in target populations with varying prevalences of significant CAD?; Given its limitation for patients with previous CABG surgery, what other patient subgroups or comorbidities were excluded from validation, and are there plans to validate its use in these populations?