HealthJOINT

HealthJOINT is an AI-powered software tool that assists in knee orthopedic surgical planning by automatically reconstructing 3D bone models and identifying anatomical landmarks from 2D X-ray images. Its methodology relies on an artificial intelligence algorithm to process DICOM radiographs, with performance studies demonstrating accurate 3D model generation (validated by RMSE against CT ground truth) and precise landmark identification for pre-operative assessment.

HealthJOINT, developed by Zebra Medical Vision Ltd., is an artificial intelligence-powered software tool designed to assist clinicians in the pre-operative planning of knee orthopedic surgeries. Its core methodology involves processing a set of 2D knee radiographs (DICOM format, including AP and LAT views) using an AI algorithm to automatically generate 3D models of the knee bones—femur, tibia, and fibula. Alongside the 3D model, the software identifies key anatomical landmarks and their precise positions. HealthJOINT supports reconstructions for both healthy bones and those affected by osteoarthritis (Kellgren-Lawrence grades 1-4). It functions as a non-user-facing component, integrating with third-party software systems via an API to deliver its output for further display and analysis, explicitly not offering clinical decisions but serving as an aid requiring professional judgment. The scientific foundation of HealthJOINT rests on its AI-driven image segmentation and 3D reconstruction capabilities. Its efficacy was substantiated in a retrospective performance study involving 67 pairs of knee X-rays and corresponding CT scans, the latter serving as the ground truth established by a board-certified radiologist. The accuracy of the generated 3D bone models was quantified using Root Mean Square Error (RMSE), yielding values of 1.14 for the femur, 1.05 for the tibia, and 0.94 for the fibula, all of which met pre-defined success criteria. Additionally, the precision of anatomical landmark identification was evaluated by measuring the distance to ground truth positions, which also satisfied performance goals. These results affirm the software's capacity to accurately convert 2D radiographic data into reliable 3D anatomical models suitable for enhancing surgical planning.

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 AI algorithm differentiate between healthy bone and various grades of osteoarthritis (Kellgren-Lawrence 1-4) during the 3D reconstruction process, and does its accuracy vary significantly across these grades?; What were the specific pre-defined success criteria for the RMSE values of the femur, tibia, and fibula, and for the anatomical landmark positioning accuracy in the performance study?; Given that the HealthJOINT software is not user-facing and integrates via an API, what are the typical computational times for generating a 3D model and landmark list from a standard set of 2D radiographs, and how might this impact clinical workflow in high-volume settings?; The performance study used CT scans as ground truth. How does the accuracy of HealthJOINT's 3D models derived from 2D X-rays compare to 3D models derived directly from other modalities like CT or MRI, particularly for complex joint deformities?; What specific anatomical landmarks are identified by the HealthJOINT software, and how was the "ground truth" for these landmarks established on the CT scans by the radiologist in the validation study?