Us2.v1

Us2.v1 is a software device utilizing machine learning for automated view classification and border detection to analyze and quantify adult transthoracic cardiac ultrasound images. It aims to support clinicians by generating comprehensive reports of cardiac structural and functional parameters, with its performance demonstrated as non-inferior to manual expert analysis in a retrospective validation study.

Us2.v1 is an image post-processing analysis software designed to assist in the diagnostic review and analysis of cardiovascular ultrasound images in DICOM format. Its primary function is to aid skilled sonographers by providing tools for manual or automated markup of echocardiographic data, which includes measurements of distance, time, area, blood flow, and quantitative analysis of cardiac function, ultimately generating a summary report for physician review. The software employs machine learning algorithms for automated view classification and border detection to facilitate this analysis, allowing sonographers to accept, modify, or reject the automated suggestions, and it organizes quantitative data against reference guidelines to support clinical decision-making for adult patients. The Us2.v1 software platform automatically processes, analyzes, and measures acquired transthoracic cardiac ultrasound images, producing a comprehensive report on numerous key cardiac structural and functional parameters. While indicated for adult patients, it has not been validated for conditions such as congenital heart disease, valve disease, pericardial disease, or intra-cardiac lesions, and its accuracy depends on good image quality and appropriate ultrasound acquisition settings. A retrospective non-interventional validation study involving 600 echocardiograms (from patients with heart failure with reduced ejection fraction and normal subjects) demonstrated that, compared to triplicate human measurements from an independent core laboratory, Us2.v1's automated measurements met a pre-specified non-inferiority margin using the individual equivalence coefficient (IEC), indicating its interchangeability with human readers for the assessed parameters.

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 machine learning architectures are utilized for the automated view classification and border detection, and what were the key characteristics of the training dataset distinct from the validation set?; How does the Us2.v1 software account for or mitigate the impact of inter-operator variability and differences in ultrasound equipment on the accuracy of its automated measurements?; Beyond the overall non-inferiority established by the Individual Equivalence Coefficient (IEC), what are the specific agreement statistics (e.g., bias, limits of agreement) for each of the individual cardiac parameters measured by Us2.v1 compared to expert human readers?; Are there plans to validate Us2.v1 for the explicitly excluded conditions such as congenital heart disease, valve disease, or pericardial disease, and what challenges are anticipated?; What mechanisms are in place to ensure ongoing performance monitoring and algorithm updates for Us2.v1 as new echocardiographic data and clinical insights emerge?