Using AI to prevent blindness
Automated Retinal Disease Assessment, or ARDA, uses artificial intelligence to help healthcare workers detect diabetic retinopathy, with future possibilities of AI algorithms to assist clinicians in identifying other diseases.
Diabetic retinopathy creates lesions in the back of the retina that can lead to total blindness. It’s important to screen people diagnosed with diabetes early, but with over 420 million people with diabetes globally, checking on every patient is an impossible task. Both the lack of awareness of the disease and the resources to screen it are huge issues.
Age of AI Series, YouTube
Copyright by Network Entertainment
Developing a diabetic retinopathy screening solution
In research published in JAMA, Google’s artificial intelligence accurately interpreted retinal scans to detect diabetic retinopathy. To do this, Google worked with a large team of ophthalmologists who helped us train the AI model by manually reviewing more than 100,000 de-identified retinal scans. This led to a development of an AI-based application called Automated Retinal Disease Assessment. This application can help doctors expand high-quality diabetic retinopathy screening programs in countries without enough eye specialists, such as India and Thailand.
Supporting 200k+ screenings and counting
Our solution is being used to detect diabetic retinopathy in India and the European Union. With almost 3k new screenings supported by ARDA weekly, we’re continuing to expand access to diabetic retinopathy screening at scale. The solution is currently being evaluated in clinical studies in the United States as well as in Thailand. We are working with multiple partners to make this solution available around the world, especially in the areas which have lower access to specialist care.
Sharing our learnings as we go
As applications of AI in medicine continue to advance, we published a paper in Nature Medicine sharing our learnings from deploying ARDA. By sharing key lessons we’ve learned from translating AI from development to deployment, we hope to help others build helpful medical AI tools that improve healthcare for everyone. Read the post.
Myths and Realities of introducing Medical AI into the real world
Quantity & Quality
High performance leads to launch
Extensive validation before launch
Just drop into workflow
Re-envision workflow Optimize care
Work is completed at launch
Continuously monitoring performance