Clarius Median Nerve AI

Clarius Median Nerve AI is a machine learning algorithm integrated into the Clarius App software that automatically segments and measures the median nerve cross-sectional area on ultrasound images acquired by Clarius ultrasound scanners. It is intended as an assistive tool for trained healthcare professionals to perform semi-automated, non-invasive measurements of the median nerve in adult patients, with users retaining responsibility for validating measurements based on clinical judgment.

Both the subject device and predicate (Clarius AI) use ultrasound image processing software implementing artificial intelligence/machine learning algorithms trained on clinical and/or artificial data. Both employ Deep Neural Networks for image segmentation and border detection, support manual adjustment of automated measurements, and are embedded in the Clarius App for iOS and Android. The primary difference is the anatomical structure: subject device measures median nerve cross-sectional area whereas the predicate measures tendon thickness (Achilles, plantar fascia, patellar tendon).

IEC 62304:2006 + A1:2015 (software lifecycle processes), ISO 14971:2019 (risk management), NEMA PS 3.1-3.20 (2022d) (DICOM), IEC 62366-1:2015 + A1:2020 (usability engineering), and ISO 15223-1:2021 (medical device symbols). Testing also followed FDA guidance on software validation, device software functions, and cybersecurity considerations.

Substantial equivalence is based on identical intended use (assistive tool for semi-automated measurement of musculoskeletal structures on ultrasound), identical principle of operation (AI/ML-based detection, segmentation, and measurement), identical technology platform (embedded in Clarius App, iOS/Android, Deep Neural Networks), and clinical performance data showing non-inferiority. The clinical verification study on 182 images from 126 subjects demonstrated the device's measurements are non-inferior to human expert measurements (p=6.497e-47, mean difference -0.065 mm²), with ICC of 0.81 versus expert reviewers. The anatomical site difference (median nerve vs. tendons) and measurement parameter difference (cross-sectional area vs. thickness) do not raise safety or effectiveness concerns given equivalent technological approach and demonstrated clinical performance.