Researchers have developed a cool new AI tool that can look at photos of the back of your eye (retina) and figure out if you’re at risk for different health problems – like eye diseases, heart disease, or even Parkinson’s disease!
The key is that the AI tool, called RETFound, was trained on a huge dataset of eye photos using a method called “self-supervised learning.” This means the researchers didn’t have to manually label every single photo by hand. Instead, the AI was able to teach itself what a normal healthy eye should look like by looking at tons and tons of photos. It’s kind of like how kids can learn to recognize cats just by seeing lots of photos of cats, without someone having to point at each photo and say “cats.”
This technique helped RETFound learn all the normal features of a healthy eye. Then the researchers could show it some labeled photos – like 100 photos of eyes from people with Parkinson’s disease and 100 photos from healthy folks. This allowed the AI to learn what makes an eye with Parkinson’s disease different. Since the AI already knew what a normal eye should look like, it could easily spot the differences.
The back of your eye (retina) can actually give clues about your overall health. That’s because the tiny blood vessels back there reflect what’s going on in blood vessels all over your body. And your eye is an extension of your brain, so it can show signs of neurological conditions like Parkinson’s too. Doctors haven’t always been great at spotting these clues, but AI tools like RETFound could help catch issues earlier.
In tests, RETFound did well at diagnosing eye diseases like diabetic retinopathy. And it was better than other AI models at predicting risk of diseases like heart disease, stroke, and Parkinson’s based on eye photos, even though it’s not perfect yet.
The researchers have made RETFound publicly available for free. This means eye doctors worldwide can potentially use and adapt it to detect diseases in their own patients. The goal is to make RETFound a helpful tool that anyone can benefit from.
Of course, there’s always a risk that flaws in the original AI could get passed on as it’s retrained for new uses. So the researchers need to be transparent about its limitations. But overall, RETFound seems like an exciting step towards using AI to improve healthcare!