BoneView

BoneView 1.1-US is a radiological computer-assisted detection software employing supervised deep learning to analyze 2D X-ray images and assist clinicians in identifying potential fractures across various anatomical regions in both adult and pediatric patients. The system processes DICOM radiographs, highlighting areas of interest with distinct confidence levels corresponding to subtle or definite fractures, and has demonstrated improved reader sensitivity and specificity in performance evaluations.

BoneView 1.1-US is an advanced software-only medical device that utilizes machine learning, specifically supervised deep learning, to assist clinicians in the detection and diagnosis of fractures from radiographs. It functions as a concurrent reading aid, analyzing DICOM images from various sources like PACS or X-ray systems, and can be deployed on-premise or in the cloud. The system processes radiographs of limbs, pelvis, rib cage, and dorsolumbar vertebrae in adults, and limbs in children/adolescents (ages 2-21), generating result files that include a summary table and annotated images highlighting potential fractures without altering the original images. These annotations distinguish between suspected (dotted line for 'DOUBT FRACT' - high sensitivity) and definite (solid line for 'FRACT' - high specificity) fractures based on the algorithm's confidence. The AI algorithm was trained on a substantial dataset of 44,649 radiographs (151,096 images). Standalone performance testing demonstrated high sensitivity and specificity for both adult and pediatric populations. For children/adolescents (2,000 radiographs), global sensitivity was 0.909 and specificity 0.821 at the high-sensitivity operating point, and sensitivity 0.792 with specificity 0.965 at the high-specificity operating point. For adults (8,918 radiographs), global sensitivity was 0.928 and specificity 0.811 at the high-sensitivity operating point, and sensitivity 0.841 with specificity 0.932 at the high-specificity operating point. A multi-reader, multi-case clinical study involving 24 readers and 480 adult cases showed that when aided by BoneView, reader diagnostic accuracy significantly improved: specificity increased by 5% (from 0.906 to 0.956) and sensitivity increased by 10.4% (from 0.648 to 0.752).

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 BoneView algorithm's performance vary across different X-ray system manufacturers and image acquisition parameters not explicitly detailed in the provided performance data?; What is the specific impact of common radiographic artifacts or suboptimal image quality (e.g., motion blur, poor patient positioning, under/over-exposure) on the BoneView algorithm's detection accuracy and the reliability of its confidence indicators (dotted vs. solid lines)?; Beyond sensitivity and specificity, what are the expected positive predictive value (PPV) and negative predictive value (NPV) of BoneView in diverse clinical populations with varying fracture prevalences for the specified anatomical regions?; How does the dual-threshold display system (DOUBT FRACT vs. FRACT) affect clinician reading times, diagnostic confidence, and the rate of false positives followed up, compared to single-threshold CAD systems or unaided reading?; Are there known limitations in detecting specific subtle fracture types, such as non-displaced hairline fractures, stress fractures, greenstick fractures in children, or fractures near complex joint articulations, and what are the corresponding performance metrics for these challenging cases?