Google ARDA eye AI crosses one million patients screened

Google's deep learning tool for diabetic retinopathy has now been used on more than a million patients across sites in India, Thailand and Australia, and the clinicians who ran those deployments have co-authored a Comment in Nature Medicine describing what the scaling actually took.

The piece is anchored by three sites: Aravind Eye Care System in Madurai, Rajavithi Hospital in Bangkok, and Lions Outback Vision at the Lions Eye Institute in Nedlands, Western Australia, alongside a Google team that includes Yossi Matias and Sunny Virmani. The tool at the center is the Automated Retinal Disease Assessment, or ARDA, which Google has publicly described as targeting "a leading and growing cause of preventable blindness." The authors argue that running one model across such different health systems "offers cross-cutting insights that may inform the expansion of healthcare artificial intelligence globally."

Google itself does not operate the clinics. It licenses ARDA to Forus Health, AuroLab and Perceptra, who deliver screenings at no cost to patients, with a stated target of six million AI-assisted screenings across India and Thailand within ten years. Two researchers on our tracker shared the Comment on the day it went live.

No per-site accuracy figures surface in the abstract preview, so the public headline is reach rather than statistical performance in each population. Google's own earlier framing is a useful anchor for how hard this part is: "the path to bringing medical AI into a real clinical environment was not easy," the company wrote in a blog post, which is roughly the gap this Comment now tries to document.