A neural network will be able to detect vision problems in premature babies.

An eleventh-grader from Novosibirsk Sergey Matveev He taught artificial intelligence to detect retinopathy—damage to the retina's blood vessels—in newborns using fundus images. Currently, diagnosing this condition requires lengthy ophthalmological and laboratory examinations, as well as an MRI of the eye sockets.

Matveyev explained that he took up this project because he wanted to simplify diagnostics for young children. First, the student studied existing retinopathy databases and then created five different neural network models. He tested each model on 20 fundus images of varying degrees of disease, and then selected the one that correctly recognized retinopathy in 98 out of 100 images.

Retinopathy in premature infants occurs because the retinal blood vessels have not yet had time to reach the periphery in the womb, and after birth, their development is impaired due to excessive or insufficient oxygen supply. Retinopathy can lead to optic nerve atrophy or even complete blindness. If detected early, the condition is treated with laser or drug injections that stop the growth of excess blood vessels in the eye.