Google Health diabetic retinopathy diagnosis

Released: November 2018

Can you improve this page?
Share your insights with us

A Google AI system capable of identifying diabetic retinopathy with 90 percent accuracy in the testing laboratory is significantly less effective in real-life. 

Developed by Google Health and tested at 11 clinics and hospitals in Thailand between November 2018 and August 2019, the system was said to perform at the equivalent level of a medical 'specialist'. 

However, Google discovered that over 20 percent of photographs taken of patients was of too low quality to be processed due to inadequate lighting or the unreliable photographic ability of the local clinic workers. 

Some nurses also discouraged patients to participate in the study because due to concerns about costs and other issues.

Operator: Ministry of Public Health, Google Health
Developer: Alphabet/Google
Country: Thailand
Sector: Health
Purpose: Identify diabetic retinopathy
Technology: Deep learning
Issue: Accuracy/reliability

Page info
Type: System
Published: May 2021
Last updated: October 2021