Clarius AI

Clarius AI is a software application utilizing non-adaptive machine learning algorithms to process musculoskeletal ultrasound images, semi-automatically placing calipers for tendon thickness measurements to aid clinicians. The system employs an AI image segmentation algorithm (U-net with MobileNetV2 backbone for training) on data from Clarius Ultrasound Scanners, providing supplementary information for foot, ankle, and knee assessments in adult patients, with its performance validated against manual clinical measurements.

Clarius AI is an advanced radiological image processing software designed for use with Clarius Ultrasound Scanner systems in musculoskeletal (MSK) applications. It employs artificial intelligence, specifically non-adaptive machine learning algorithms (including U-net with MobileNetV2 for model training and an Exception U-Net for operations), to assist trained healthcare professionals in the non-invasive measurement of tendon thickness. The software analyzes ultrasound images in real-time, segmenting relevant tendons such as the plantar fascia, Achilles’ tendon, and patellar tendon in adult patients. It then semi-automatically places adjustable calipers at the thickest region of the segmented tendon, providing an overlay on the ultrasound image and allowing users to make manual adjustments, thereby offering supplementary information to inform clinical management without replacing clinical decision-making. The performance of Clarius AI (MSK model) was rigorously evaluated through non-clinical and clinical testing. AI model training involved a dataset of over 20,000 ultrasound images, and validation with data from 73 subjects demonstrated high accuracy, achieving a 96% average Dice score and 94% mean Intersection over Union (IoU) for tendon segmentation. Verification testing confirmed that the software's automated tendon thickness measurements were non-inferior to manual measurements performed by expert clinicians, with a mean difference between auto-measurements and mean manual measurements of 0.03% (p-value 9.0 x 10^-7) within clinically significant margins. Design validation studies further established that Clarius AI is clinically usable and meets user needs for automatic segmentation highlighting, semi-automatic caliper placement with manual adjustment, and persistent storage in MSK ultrasound applications.

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 performance of the Clarius AI (MSK model) vary across different patient demographics (e.g., age, BMI, presence of specific pathologies) not explicitly detailed in the training and validation datasets?; What is the typical learning curve for clinicians to effectively utilize Clarius AI and integrate its semi-automatic measurements, including manual adjustments, into their existing MSK ultrasound workflow?; Could the AI model's performance be significantly affected by variations in ultrasound machine settings or operator skill levels beyond the Clarius L7 and L15 transducers and trained users involved in the study?; What are the specific failure modes or limitations of the AI algorithm in tendon segmentation and caliper placement, and what is the reported frequency of instances requiring substantial manual correction by the user?; Given the 'non-adaptive' nature of the machine learning algorithm, what is the anticipated strategy for updating or re-validating the Clarius AI model to maintain optimal performance as new ultrasound imaging techniques emerge or the understanding of tendon pathology evolves?