IDx-DR v2.3

IDx-DR v2.3 is an artificial intelligence-powered software designed to automatically detect more than mild diabetic retinopathy from fundus images captured by the Topcon NW400 camera. The system employs an AI algorithm within its Analysis component to analyze retinal images for pathologic features indicative of mtmDR, providing healthcare providers with a screening result to guide patient referral or rescreening decisions.

IDx-DR v2.3 is an autonomous artificial intelligence (AI)-based software system designed for healthcare providers to automatically detect more than mild diabetic retinopathy (mtmDR) in adults with diabetes who have not been previously diagnosed with the condition. The system utilizes the Topcon NW400 fundus camera to capture two retinal images per eye (macula- and disc-centered), which are then processed by the IDx-DR Analysis software; this software employs AI algorithms to assess image quality and identify the presence or absence of mtmDR based on specific pathologic features from these images. The overall system architecture includes a client application at the user site for image acquisition and result display, a central IDx-DR Service for secure data handling, communication, and cybersecurity, and the core AI-driven IDx-DR Analysis System, all operating without requiring patient information beyond the fundus images. This version, IDx-DR v2.3, introduces modifications primarily to its software components, including an updated image quality classifier and improved processing speed in the IDx-DR Analysis System, and enhanced user interface feedback in the IDx-DR Client. Its clinical performance was evaluated in a retrospective study using data from a prior version's pivotal trial, analyzing images from 850 participants. For final submission images (acquired with as-needed pharmacologic pupil dilation), IDx-DR v2.3 demonstrated a diagnosability rate of 95.2%, a sensitivity of 87.69% (95% CI: 82.24%-91.95%), and a specificity of 90.07% (95% CI: 87.42%-92.32%) in detecting mtmDR among diagnosable participants, with results generally comparable to the predicate device.

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 the IDx-DR v2.3 image quality classifier compare to the previous version (v2.1.1) in terms of specific image artifact rejection and acceptance rates?; What are the computational requirements (e.g., processing time, memory usage) for the IDx-DR Analysis System v2.3.0, and how does its improved speed impact the clinical workflow?; Given the slight decrease in diagnosability for final submission images (95.2% for v2.3 vs. 96.4% for v2.0), what factors contributed to this change, and were there specific image characteristics that v2.3 found more challenging to analyze?; What specific "pathologic features" does the AI algorithm in IDx-DR Analysis System v2.3.0 prioritize or weigh for detecting mtmDR, and how robust is it to variations in image quality that are still deemed 'sufficient'?; How does the system handle edge cases or atypical presentations of diabetic retinopathy, and what is the protocol if the AI encounters an image it cannot confidently classify but does not flag as "insufficient quality"?