At St. Petersburg Electrotechnical University "LETI", a neural network has been trained to detect retinal pathology in humans that occurs in diabetes, even in its early stages. This is diabetic retinopathy, which may be asymptomatic at first, but can then lead to vision loss.
The neural network detects this disease in its early stages with an accuracy of 88.7% based on images of the retina. It relies on hemorrhages, accumulation of fats, growth of blood vessels, and thinning of their walls on the retina. As the creators of the project shared with Gazeta.Ru, the neural network can also diagnose diabetes - diabetic retinopathy occurs due to a violation of normal blood glucose levels.
In the future, it will be possible to develop a device in the form of glasses that can automatically scan the eye, finding the region of the retina using AI, and also assess not only the stage of diabetic retinopathy, but also the stage of other eye diseases associated with diabetes: cataracts, glaucoma.
Russian doctors use AI not only for diagnostics, but also for operations - recently, the first operation in Russia with its participation was successfully performed. Removal of a malignant kidney tumor through a puncture of the abdominal wall was helped by the domestic Razum system.