EyeArt

The EyeArt system is an artificial intelligence-based software designed to automatically detect more than mild and vision-threatening diabetic retinopathy from retinal fundus images, intended for use by healthcare providers in both primary care and eye care settings. Its methodology relies on deep learning algorithms analyzing dual-field images per eye captured by specific Canon cameras, with clinical trials demonstrating high sensitivity, specificity, and imageability for detecting different stages of the disease.

EyeArt is an artificial intelligence-powered software as a medical device designed for healthcare providers to automatically detect specific levels of diabetic retinopathy in adults diagnosed with diabetes who have not been previously diagnosed with more than mild diabetic retinopathy. The system utilizes an ensemble of deep learning algorithms to analyze retinal fundus images captured by Canon CR-2 AF or Canon CR-2 Plus AF cameras. For each eye, two 45-degree field-of-view color images (optic nerve head centered and macula centered, with a resolution of 1.69 megapixels or higher) are transmitted from a client software on the operator's computer to a remote Analysis Computation Engine via a secure server. This engine employs machine learning to identify patterns indicative of 'more than mild diabetic retinopathy' (mtmDR), defined as moderate non-proliferative diabetic retinopathy or worse and/or diabetic macular edema, and 'vision-threatening diabetic retinopathy' (vtDR), defined as severe non-proliferative or proliferative diabetic retinopathy and/or diabetic macular edema, based on the International Clinical Diabetic Retinopathy severity scale. The EyeArt system was evaluated in extensive non-clinical and clinical testing, including a prospective, multi-center pivotal trial with an analysis population of 655 participants across U.S. primary care and general ophthalmology centers. Its performance was assessed against expert grading from the Fundus Photography Reading Center using the ETDRS scale. In sequentially enrolled cohorts, EyeArt demonstrated high sensitivity (e.g., 100% for mtmDR in primary care, 94.9% for mtmDR in ophthalmology) and specificity (e.g., 92.0% for mtmDR in primary care, 86.7% for mtmDR in ophthalmology), with similarly strong performance for vtDR detection; these results were largely consistent in an enrichment-permitted cohort designed to include more severe cases. The system achieved high imageability (96.5% or greater), with a significant portion of gradable results obtained on the first attempt without dilation. Separate precision studies confirmed high intra-operator repeatability (overall agreement >93%) and inter-operator reproducibility (overall agreement >90%) for both mtmDR and vtDR outputs, and human factors validation indicated that operators with no prior retinal photography experience can be effectively trained to capture adequate quality images.

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 EyeArt algorithm's performance vary across different severities within the 'more than mild' and 'vision-threatening' diabetic retinopathy categories, beyond the binary detection?; What are the specific image quality parameters that lead to an 'ungradable' EyeArt result, and what is the typical workflow or recommendation for these cases in a clinical setting?; Given the use of both sequential and enriched enrollment cohorts in the clinical trial, how might the differing disease prevalence in these cohorts have influenced the reported sensitivity and specificity metrics, particularly in primary care versus ophthalmology settings?; The human factors study indicated successful training for operators with no prior retinal photography experience; what was the duration and nature of this training, and how does it compare to training requirements for similar ophthalmic imaging devices?; How does the performance of EyeArt, particularly for vision-threatening diabetic retinopathy detection, compare with the predicate device IDx-DR, which was indicated only for more than mild diabetic retinopathy, and what are the clinical implications of this expanded detection capability?