EFAI Bonesuite XR Bone Age Pro Assessment System (BAP-XR-100)

The EFAI Bonesuite XR Bone Age Pro Assessment System (BAP-XR-100) is a software medical device that employs deep learning techniques to analyze left-hand radiographs according to the Greulich-Pyle method, providing pediatric radiologists with a quantitative bone age assessment for patients aged 2 to 16. Its clinical utility as an adjunct diagnostic tool is supported by performance studies demonstrating high accuracy and consistency compared to expert radiologist consensus across diverse patient demographics and imaging conditions.

The EFAI Bonesuite XR Bone Age Pro Assessment System (BAP-XR-100) is a software tool designed to assist pediatric radiologists in quantifying bone age for patients aged 2 to 16 years. It utilizes deep learning algorithms to analyze 2D Posterior Anterior (PA) views of left-hand radiographs, referencing the Greulich-Pyle (GP) method for bone age estimation. The system outputs the estimated bone age in years as a standard JSON message for integration with existing Picture Archiving and Communication Systems (PACS), Radiology Information Systems (RIS), or Hospital Information Systems (HIS), and is intended as an adjunct tool, meaning its results should not solely determine diagnostic decisions; it requires input images to be clear of major bone destruction, deformity, fractures, or significant artifacts. The system's performance was validated through extensive nonclinical and clinical testing. Nonclinical evaluation involved a dataset of 26,222 cases, where Deming regression analysis demonstrated a strong correlation (intercept -0.06, slope 1.00) between the EFAI BAPXR output and ground truth established by radiologists. Clinical validation utilized 600 diverse U.S. cases (ages 2-16), independent of the training data, with ground truth determined by four U.S. board-certified pediatric radiologists. The primary endpoint, Deming regression between EFAI BAPXR output and ground truth, was successfully met (intercept -0.07, 95% CI [-0.13, -0.01]; slope 1.00, 95% CI [0.99, 1.00]). Performance data indicated that over 88% of cases had a difference of less than 0.5 years from the ground truth, and a Bland-Altman plot showed high consistency with 95% limits of agreement ranging from -0.517 to 0.743 years. Subgroup analyses further confirmed consistent and reliable performance across various genders, ages, races, clinical organizations, and X-ray equipment.

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 system's performance vary in the 'Late-puberty' subgroup, where the Deming regression intercept was -1.10 and slope 1.07, compared to other age groups, and what might account for this difference?; What specific deep learning architecture is employed by the EFAI BAPXR, and how was the Greulich-Pyle method computationally interpreted and implemented for feature extraction and age estimation?; Given the requirement that images be without 'major bone destruction, deformity, fracture, excessive motion, or other major artifacts,' does the system include an automated pre-screening or quality assessment module to flag unsuitable images?; How does the system's bone age assessment time and consistency compare to manual Greulich-Pyle assessments performed by individual pediatric radiologists, considering inter-rater and intra-rater variability?; Beyond the provided subgroup analysis for X-ray manufacturers, what specific image characteristics (e.g., resolution, noise levels, contrast) influenced by different scanners have the most impact on the algorithm's accuracy, and are there calibration recommendations?