AEYE-DS

AEYE-DS is an artificial intelligence-based software device intended for use by healthcare providers to automatically screen for more than mild diabetic retinopathy (mtmDR) in diabetic adults using fundus images from a Topcon NW400 camera. Its methodology relies on AI analysis of retinal images, with clinical studies demonstrating high sensitivity and specificity, supporting its utility for detecting mtmDR.

The AEYE-DS is a retinal diagnostic software device utilizing artificial intelligence 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 operates with a Topcon NW400 fundus camera, where a local client module captures retinal images (either one or two per eye) and transmits them via a secure internet connection to a remote server. The server hosts service and analytics modules that process these images to identify disease features, returning a diagnostic result (mtmDR detected, not detected, or insufficient quality) to the client for healthcare provider review. The software architecture includes components for client interaction, service processing, analytics, reporting/archiving, and system security. A pivotal clinical study involving 531 subjects across 8 sites demonstrated the device's performance. When using one macula-centered image per eye, the AEYE-DS showed a sensitivity of 92.98% and specificity of 91.36% for mtmDR detection. With two images per eye (macula-centered and optic disc-centered), sensitivity was 94.74% and specificity was 88.64%. Imageability was high at over 99%, with most subjects not requiring pupil dilation. Precision studies showed high intra-operator, between-operator, and between-device reproducibility. A human factors validation study confirmed that novice operators found the device easy to use after initial training, successfully submitting images and obtaining diagnostic reports.

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 AEYE-DS algorithm specifically differentiate features of 'more than mild diabetic retinopathy' from milder forms or normal retinal presentations?; What was the rate of 'insufficient quality' images initially, and how did this change with the maximum of 6 image submission attempts by novice operators?; Given the slightly lower specificity with the 2-images-per-eye protocol (88.6%) compared to the 1-image-per-eye protocol (91.4%), what are the clinical considerations or trade-offs when deciding which protocol to use?; How does the performance of the AEYE-DS device vary with different demographic subgroups beyond those analyzed (e.g., specific comorbidities, duration of diabetes categories not tested)?; What are the specific cybersecurity measures implemented in the AEYE-DS system, particularly concerning the transfer and storage of patient fundus images on remote servers?