Accurate measurement of geographic atrophy (GA) lesion area is a central requirement for clinical trial enrollment, yet many sites lack a practical method to determine eligibility at the point of care. The result is frequent screen failures, increasing burden on patients, investigators, and reading centers tasked with confirmatory review.

A real-world study presented at the 2026 Association for Research in Vision and Ophthalmology (ARVO) meeting in Denver found that use of a clinic-based artificial intelligence (AI) tool reduced screen failure rates and showed close agreement with manual measurements in GA trial prescreening. Investigators from the Wisconsin Reading Center at the University of Wisconsin-Madison evaluated an algorithm that measures GA area from fundus autofluorescence images in real time.

Retina specialists prospectively enrolled patients with unifocal or multifocal GA meeting prespecified lesion size criteria. Among 107 eyes from 62 participants, AI-assisted prescreening identified 80 eyes as eligible and 27 as ineligible. A reading center confirmed eligibility in 67 of the 80 prescreened eyes, while 1 of 27 deemed ineligible by AI was later found eligible.

Mean GA area was 9.39 mm² by manual planimetry and 8.98 mm² by AI. False-positive and false-negative rates were 12% and 1%, respectively, with discrepancies attributed to measurement error and confounding lesions. The screen failure rate was 16% with AI-assisted prescreening, compared with 25% using standard reading center workflows, suggesting a potential role for clinic-based tools in improving trial efficiency.

These findings suggest that clinic-based AI tools may reduce screening burden and improve trial efficiency, observed Amitha Domalpally, MD, senior author on the poster. RP