Researchers at Stanford have devised an eye test that can be done on a home computer and uses AI to improve its accuracy.

The generations-old eye test that uses a chart full of alphabets of varying sizes or even the new-fangled digital eye tests may soon be disrupted. The virus pandemic has forced shelter-in-place and homebound people to seek an alternative. What if your glasses broke and you didn’t have a spare set? There may soon be an AI-based alternative. It’s an eye test that can be done online, with the patient using a computer at home. (RTInsights.com)

Stanford University research

Chris Piech, Ali Malik, Robert T. Chang, and Charles Lin of Stanford University, and Laura M. Scott of Mapstone.org have collaborated on a study titled “The Stanford Acuity Test: A Precise Vision Test Using Bayesian Techniques and a Discovery in Human Visual Response.”

The paper emphasizes that traditional chart tests suffer from high variance in their results due to the patient guessing the letters. There is also an “approximation error due to the need to discretize letter sizes on a chart.”

Meanwhile, digital eye tests are an improvement over charts, they do not dramatically improve accuracy over the latter.

The algorithm

The researchers devised an AI algorithm named the Stanford Acuity Test (StAT). It more accurately measures visual acuity by merging the better features of both charts and digital exams.

“The result is a test that is 74% more accurate than the analog chart,” said the paper. “Compared to the previous best digital exam, our experiments show an error reduction of up to 67%.”

“For patients with more serious eye disease, the novel ability to finely measure acuity from home could play a crucial role in early diagnosis and effective treatment,” it added.

The paper invites interested persons to test the algorithm (screenshot below) by taking an online eye test at myeyes.ai. (Disclaimer: This test is a mathematical demo. It has not been FDA approved. Do not use it as a substitute for a medical opinion.)

Conclusion

“This algorithm outperforms all prior approaches for this task while also providing reliable, calibrated notions of uncertainty for its final acuity prediction,” say the Stanford researchers. “Our approach is incredibly accurate, easy to implement, and can even be used at home on a computer.”

Related Story:     South Korea’s SK Telecom Deploys Robotic COVID-Watcher