IDx-DR

IDx-DR is an artificial intelligence-powered software designed to automatically detect more than mild diabetic retinopathy from fundus images acquired using a Topcon NW400 camera, aiming to facilitate early screening in diabetic adults. The system employs a client-server architecture for image acquisition, secure transfer, AI-based analysis of retinal features, and provides actionable diagnostic outputs for healthcare providers based solely on image analysis.

IDx-DR is an artificial intelligence-based software medical device designed for healthcare providers to automatically detect more than mild diabetic retinopathy (mtmDR) in adults diagnosed with diabetes who have not previously been diagnosed with the condition. The system utilizes a Topcon NW400 fundus camera to capture two specific retinal images per eye (macula-centered and disc-centered, 45° field of view, requiring at least 22 pixels per degree on the retina). These images are processed by the IDx-DR Analysis software, which employs AI algorithms to assess image quality and identify the presence or absence of mtmDR. The system architecture comprises an IDx-DR Client for guided image acquisition and results display, IDx-Service for secure data transfer and management via a webserver, and the core IDx-DR Analysis on dedicated servers for image processing, requiring only fundus images for its analysis. The underlying science involves AI technology analyzing specific disease features from fundus images. This version of IDx-DR introduces enhancements such as DICOM image submission, improved client feedback (distinguishing between "submission not found" and "submission not ready"), image analysis statistics tracking, a guided user workflow with in-exam image quality feedback and re-submission capability, local image retention, and a training mode. The system's output provides a clear diagnostic screening result: "mtmDR detected" (recommending referral to an eye care professional), "mtmDR not detected" (recommending rescreening in 12 months), or "Insufficient image quality." Comprehensive non-clinical testing confirmed software safety, effectiveness, cybersecurity, and usability, including studies demonstrating that untrained operators can capture quality images using the guided workflow. A protocol is also established for managing future algorithm changes to maintain safety and efficacy.

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. What are the specific image quality metrics (e.g., resolution, clarity, field of view coverage) that IDx-DR Analysis uses to determine "insufficient image quality" versus acceptable for analysis?; How does the "training mode" within the IDx-DR Client prepare new users, and what metrics are used to assess operator competency before live patient use?; What are the specific AI architectural details or types of neural networks employed in IDx-DR Analysis for feature detection of diabetic retinopathy?; Considering the device outputs a referral decision or rescreening interval, how does the system account for variations in diabetes severity or patient risk factors beyond the retinal images themselves, given no other information is required?; What is the process and criteria within the "protocol to mitigate the risk of algorithm changes" for determining when an algorithm update requires new regulatory scrutiny versus internal validation only?