TOPLINE

An AI-based system (iPredict-AMD) demonstrated strong performance in detecting referable age-related macular degeneration (AMD) in adults older than 50 years in primary care and general ophthalmology settings.
 

METHODOLOGY

• A prospective study was conducted across primary care and general ophthalmology clinics in New York City to validate an AI-based screening tool for detecting referable AMD.
• A total of 845 adults older than 50 years without prior AMD diagnosis were recruited and underwent non-dilated fundus photography of both eyes. Overall, 696 adults (median age, 59 years; 373 women) completed the study.
• The AI tool was built using five deep learning models trained on over 116,000 fundus images from more than 4000 participants in the AREDS study, and it classified each image into one of four AMD stages before combining the results into a single score.
• Researchers captured non-dilated retinal photos using an automated fundus camera and ran them through the AI system, while dilated images from the same patients were graded by ophthalmologists as the gold standard for comparison.
• The AI system's performance was evaluated on both per-patient and per-eye bases, with metrics including area under the curve (AUC), sensitivity, and specificity.

TAKEAWAY

• For identifying more-than-early AMD at the patient level, the AI system achieved an AUC of 0.92. It correctly detected 90.27% of patients who truly had the condition (sensitivity) and correctly ruled out 83.36% of those who did not have the disease (specificity).
• At the individual eye level, the tool showed similar performance, with an AUC of 0.91.
• The tool correctly identified 102 out of 113 participants with referable AMD, missing 11 borderline cases who were advised to return for yearly screening.
• More than 97% of patients who screened negative for AMD truly did not have the disease, based on results at both the patient level and eye level.

IN PRACTICE

"The proposed AMD analysis tool …. can be safely and rapidly deployed in a broad range of clinical settings, thus being an effective tool for AMD screening and blindness prevention," the authors of the study wrote.

SOURCE

The study was led by Alauddin Bhuiyan, iHealthscreen Inc, New York City. The study was published online on July 29 in Scientific Reports.

LIMITATIONS

Many participants dropped out from the study. The study was conducted at a single location.
 

DISCLOSURES

The study received funding from the National Institutes of Health SBIR.  One author reported holding a patent related to an image-based screening system for predicting individuals at risk of late-stage AMD.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.