Detecting Retinopathy of Prematurity Disease Based on Fundus Image Dataset

Abstract

Retinopathy of prematurity (ROP) is a condition that has the potential to cause blindness in infants. ROP detection is an imminent necessity, and it appears to be a trustworthy, dependable, and cost-effective addition to humans. ROP is result of pathological neovascularization. It is defined by immature early embryogenesis of the vascular system of the retina. Early detection and therapy of ROP, on the other hand, can considerably enhance the high-risk individual’s vision acuity. As a consequence, detecting ROP early is crucial for preventing vision impairment. ROP occurs in premature newborns born at 32 weeks and with a low weight at birth (less than or equal to 1.5 kg). It is estimated that 19 million children worldwide suffer from vision impairment. Over 1.84 million of the children were anticipated to have experienced ROP. ROP causes around 11% of people to become entirely blind or severely visually impaired, with the other 7% experiencing mild to moderate visual impairment. To identify ROP in babies, it used neural networks such as Inception-V3, ResNet-50, and VGG-19. This model aids in assessing whether a premature newborn has ROP or not. It also classified the disease’s severity and predicted accuracy. We obtained the highest accuracy of 99% for ResNet-50 among the 3 algorithms.