Experts from the University of Oregon science and health (USA) has developed a unique method of identifying the pathologies associated with ocular blood vessels and contributing to the development of blindness in premature infants.
A new method based on convolutional neural networks, when testing showed 98% accuracy in the diagnosis of abnormalities and surpassed in this professional ophthalmologists.
Retinopathy of prematurity, manifested in children born before term with a not to exceed 1250 g weight is one of the main causes of vision loss in children, said experts. The detection of plus disease at an early stage in the incubation period of the child and the provision of timely therapy pathology disappears in a few months. Otherwise, running disease contributes to retinal detachment and contributes to the development of blindness.
Doctors to detect vision problems in preterm guided by the comparison of patients with ocular blood vessels healthy, but often the analysis was incorrect. American experts proposed to diagnose possible pathology using the method of machine learning. The algorithm described in the pages of JAMA Opthalmology, works on the principle of convolutional neural networks. Taught the car for over 5500 frames with images of healthy retinas and babies with problems ranging from the development of plus disease and to its progression.
When testing a new way he has shown accuracy of up to 98% in the detection of the disease. This method exceeded the performance diagnosis conducted by health professionals who study the problem of vision loss in premature babies for over 10 years. The developers of the algorithm believe that it will reduce the likelihood of misdiagnosis in the study of the retina to a minimum, and this will allow you to save the eyesight of many children.