AI Platform 2.0 (AIP002)

The AI Platform 2.0 is an advanced software medical device that employs deep convolutional neural networks to process adult cardiac and lung ultrasound images, automating the detection, measurement, and quantification of key clinical parameters. Its purpose is to enhance diagnostic workflows and image assessment by providing quantitative data like LVEF, myocardial wall thickness, IVC diameter, A/B lines, and real-time image quality feedback to healthcare professionals.

Exo AI Platform 2.0 (AIP002) is a Software as a Medical Device (SaMD) designed to aid qualified healthcare providers in the image-based assessment of adult ultrasound examinations. It processes DICOM-format ultrasound images using deep convolutional neural networks for segmentation, landmark detection, and classification to simplify workflows by evaluating, quantifying, and generating reports on medical parameters. The platform features Lung AI for identifying A-lines and B-lines and Cardiac AI for quantifying Left Ventricular Ejection Fraction (LVEF), myocardium wall thickness (Interventricular Septum and Posterior wall), and Inferior Vena Cava (IVC) diameter. Additionally, both AI modules provide real-time per-frame and per-clip quality scores for lung scans and specific cardiac views (Apical and PLAX), with its AI modules also available as accessories for integration into other ultrasound systems. The performance of AI Platform 2.0's new features was validated using diverse, independent test data. For left ventricle wall thickness and IVC measurements on 100 subjects, Inter-class Correlation (ICC) with expert ground truth was high: 0.93 for Interventricular Septum, 0.94 for Posterior Wall, 0.93 for IVC Dmin, and 0.94 for IVC Dmax. The Quality AI, validated retrospectively on 184 patients (226 clips), showed an ICC of 0.94 against sonographer ratings. Real-time use by 26 users (including novices) on 396 scans demonstrated that 98.3% of expert-rated high-quality clips were also rated diagnostically adequate by the AI, and 98.0% of AI-rated adequate scans were deemed diagnostic by experts. A Predetermined Change Control Plan (PCCP) is in place for specified future software modifications, ensuring continued safety and effectiveness. Users are cautioned that the platform's analysis should not be the sole basis for patient management decisions.

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 'Quality AI' algorithm differentiate between various image artifacts and true anatomical features to assign a quality score, especially in scans performed by novice users?; What specific ultrasound scanner models and manufacturers is the AI Platform 2.0 currently validated for, and what is the process for extending compatibility to new devices?; Given the Predetermined Change Control Plan, what are the defined non-inferiority margins for performance metrics when new training data or architectural modifications are introduced, particularly for critical measurements like LVEF?; How does the AI Platform 2.0 handle ultrasound images with significant anatomical variations or rare pathologies not extensively represented in its training dataset, and what is the reported failure rate or confidence scoring in such scenarios?; What is the computational overhead for the real-time quality score feedback, and how does it impact the performance of the host ultrasound system, especially portable devices?