RBknee

RBknee is an automated software device utilizing machine learning and computer vision to analyze knee radiographs, providing objective measurements of joint space width and assessments of osteoarthritic features according to OARSI and Kellgren-Lawrence criteria. Its purpose is to aid medical professionals by offering standardized, quantitative data derived from CR or DX images to support the evaluation of knee osteoarthritis.

RBknee is a fully automated image processing software designed to assist medical professionals in evaluating knee radiographs for signs of osteoarthritis (OA). It processes computed (CR) or directly digital (DX) images, specifically standing, fixed-flexion anterior-posterior (AP) and posterior-anterior (PA) views of the knee. The software employs computer vision and machine learning algorithms, trained on a substantial dataset of medical images, to automatically analyze these radiographs. Its primary functions include measuring the minimum joint space width (JSW) and assessing the presence or absence of key osteoarthritic features such as sclerosis, joint space narrowing, and osteophytes, based on established OARSI (Osteoarthritis Research Society International) criteria. The device generates both visual and text reports. The visual report includes the original radiograph with overlays highlighting the minimum JSW and a table summarizing JSW measurements in millimeters, OARSI gradings for joint space narrowing, osteophytes, and sclerosis, and the Kellgren-Lawrence (KL) score for radiographic knee OA. The text report provides a narrative summary of these findings and an overall impression based on the KL score. RBknee does not interpret results but provides objective measurements and classifications. Clinical performance validation using the Osteoarthritis Initiative (OAI) database showed high sensitivity and specificity for detecting Kellgren-Lawrence status (0.88 sensitivity, 0.87 specificity), joint space narrowing (0.82 sensitivity, 0.87 specificity), osteophytosis (0.89 sensitivity, 0.78 specificity), and subchondral sclerosis (0.84 sensitivity, 0.87 specificity). JSW measurements demonstrated strong agreement with OAI data, with slopes near 1.00 and intercepts near zero for both medial and lateral compartments.

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 RBknee's machine learning algorithms vary across different patient demographics (e.g., age, gender, BMI) not explicitly detailed in the OAI subset used for validation?; What was the inter-rater reliability of the ground truth OARSI and Kellgren-Lawrence grading performed by the two physicians with adjudication, and how might this variability impact the reported performance metrics of RBknee?; Can RBknee differentiate between early and advanced stages of specific OARSI features (e.g., osteophyte size, severity of sclerosis) beyond binary presence/absence, and if so, what is its accuracy for finer gradations?; What specific image features does the machine learning model prioritize when assessing for osteophytes, and how robust is it to variations in radiographic quality or technique?; How does RBknee's automated Joint Space Width measurement compare to manual or semi-automated methods in terms of both accuracy and time efficiency in a real-world clinical workflow?