Score a win for AI in its quest to replace all of our jobs, including doctors’.
Scientists have developed a machine learning app that is being used to detect early symptoms of eye disease by scanning through a collection of photos. The app, called White Eye Detector, uses an algorithm to look for signs of white eye reflections in photos pretty similar to the red eyes you sometimes see in photos. The app is so accurate that in some children it spotted symptoms up over a year before it was diagnosed by doctors.
White Eye Detector was trained by researchers using 53,000 photos of 40 children, half of whom had been diagnosed with an eye disease. Seeing a red spot on eyes is a sign that they’re healthy. White eyes can indicate retinoblastoma (cancer of the retina), retinopathy of prematurity, cataracts, or Coats Disease.
Retinoblastoma develops tumors rapidity. Within six months to a year, the tumor can break up and metastasize down your optic nerve to the brain and kill you. Because retinoblastoma only occurs in children it can be hard to detect because they don’t have a way to clearly communicate when something is wrong with their eye.
Bryan Shaw, a professor at Baylor University, and his team developed the app after Shaw’s own son was diagnosed with retinoblastoma when he was four months old. Going back through early baby pictures, the software found signs of white eye when he was just 12 days old.
Here’s a sample image of what the apps looks for:
Photo: Michael Munson/Science Advances
Parents love taking pictures of their newborn which provides plenty of data for the app’s algorithm to churn through and provide an early warning. In a recent study using White Eye Detector, 16 of 20 children with retinoblastoma were spotted with symptoms on an average of 1.3 years before doctors detected it. The free app is available on iOS and Android and while it can’t diagnose retinoblastoma, it can detect white eyes which you should have checked out by a physician.
