AI Platform (AIP001)

The AI Platform (AIP001) is a Software as a Medical Device designed for noninvasive processing of adult ultrasound images, employing Deep Convolutional Neural Networks to detect, measure, and calculate medical parameters related to cardiac and lung function. Its methodology involves analyzing DICOM ultrasound images with specific AI tools for LVEF quantification and lung artifact detection, with clinical performance validated against expert consensus across diverse patient populations and imaging conditions.

The AI Platform (AIP001) is a software as a medical device (SaMD) intended for the noninvasive processing of ultrasound images to assist qualified users in the image-based assessment of adult patients with suspected disease. It processes imported Digital Imaging and Communications in Medicine (DICOM) images from a specific range of ultrasound scanners, enabling users to detect, measure, and calculate relevant medical parameters of structures and function, primarily focusing on the heart and lungs. Key features include 'LVEF AI' for AI-assisted quantification of ejection fraction from cardiac ultrasound images, and 'Lung AI' for AI-assisted suggestion of lung structures and artifacts like A-lines and B-lines. The platform is designed to simplify workflow, help trained healthcare providers evaluate, quantify, and generate reports, but patient management decisions should not be based solely on its analysis, and manual adjustments by users are permitted. The device's performance was validated using clinical data distinct from its training datasets. For cardiac assessment (LVEF AI), evaluation on 151 subjects showed high reliability, with an overall intraclass correlation coefficient (ICC) of 0.93 and a root mean square difference (RMSD) of 5.90 when compared to expert consensus on ejection fraction measurements across various views (Parasternal Long-axis, Apical Biplane, AP4, AP2). For lung assessment (Lung AI), evaluation on 125 subjects demonstrated strong agreement for A-lines (Cohen's kappa = 0.84) and high reliability for B-line counts (ICC = 0.97) compared to expert consensus. The validation encompassed diverse demographic variables (age, BMI, ethnicity), various ultrasound devices, and common clinical confounders, showing consistent performance.

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 AI Platform's performance in LVEF quantification (ICC and RMSD) vary when using images from ultrasound devices with frequencies outside the 1.2-4 MHz range tested?; What specific training and expertise level is assumed for 'trained healthcare providers' to effectively use the AI Platform and interpret its outputs, particularly when manual adjustments are made?; How does the 'Lung AI' differentiate between various causes of B-lines (e.g., pulmonary edema vs. fibrosis) or are there plans to enhance this specificity?; What were the specific performance metrics (e.g., sensitivity, specificity) for the AI Platform in identifying A-lines and B-lines across the diverse demographic subgroups and clinical confounders mentioned, beyond the overall Kappa and ICC values?; Can the AI Platform integrate with various existing PACS or EMR systems, and what are the technical requirements for such integrations beyond DICOM compatibility?