DeepXray

DeepXray is an AI-driven software designed to assist medical professionals by automatically analyzing knee radiographs to quantify joint space width and femoral-tibial angle, and assess osteoarthritis features using Kellgren-Lawrence and OARSI criteria. Its methodology relies on computer vision algorithms trained on a large dataset, with outputs presented in an interactive web report to support, but not replace, clinical judgment in diagnosing knee osteoarthritis.

DeepXray is a standalone software device employing artificial intelligence and computer vision algorithms to assist trained medical professionals, such as radiologists and orthopedists, in analyzing knee radiographs. It processes either computed (CR) or directly digital (DX) posterior-anterior or anterior-posterior knee images from standing, fixed-flexion views to automatically detect and measure radiographic abnormalities associated with knee osteoarthritis (OA), with the aim of aiding diagnosis without replacing comprehensive patient evaluation or being solely relied upon for diagnostic confirmation. DeepXray provides quantitative measurements including minimum joint space width and the femoral-tibial angle, alongside qualitative assessments of knee OA severity using the Kellgren-Lawrence Grade, and specific features like joint space narrowing, osteophytes, and sclerosis based on OARSI criteria. Its output is presented as an editable web-based summary report, which includes markup images, quality warnings, and an interface for clinicians to review and modify AI-generated results. Performance validation on a large, independent dataset from the Osteoarthritis Initiative (OAI) demonstrated good agreement with expert clinician assessments, achieving sensitivities for various OA indicators between 0.84 and 0.88 and specificities between 0.80 and 0.88, and high accuracy for quantitative measurements, confirming its utility as an adjunctive tool.

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 DeepXray's performance vary across different severities of knee osteoarthritis (e.g., early vs. advanced stages) as defined by Kellgren-Lawrence grades?; What is the impact of the 0.3% of test samples blocked by the automatic quality control mechanism on the overall utility and potential biases of the system in a real-world clinical workflow?; How does the editable web report feature, allowing clinician modification, affect inter-observer variability and the final diagnostic accuracy compared to using the raw AI output alone?; Given the training dataset included images from various X-ray manufacturers, how robust is DeepXray's performance when encountering images from manufacturers not extensively represented or entirely absent in the training data?; What is the typical time saved, if any, by clinicians using DeepXray for knee OA assessment compared to traditional manual review methods, and how does this integrate into existing radiological workflows?