Applications of Artificial Intelligence for Retinopathy of Prematurity Screening.

摘要: OBJECTIVES: Childhood blindness from retinopathy of prematurity (ROP) is increasing as a result improvements in neonatal care worldwide. We evaluate the effectiveness artificial intelligence (AI)–based screening an Indian ROP telemedicine program and whether differences severity between units (NCUs) identified by using AI are related to oxygen-titrating capability. METHODS: External validation study existing AI-based quantitative scale for on data set images Retinopathy Prematurity Eradication Save Our Sight India. All were assigned score (1–9) Imaging Informatics Deep Learning system. calculated area under receiver operating characteristic curve sensitivity specificity treatment-requiring prematurity. Using multivariable linear regression, we evaluated mean median each NCU function birth weight, gestational age, presence oxygen blenders pulse oxygenation monitors. RESULTS: The detection was 0.98, with 100% 78% specificity. found higher (interquartile range) NCUs without monitors, most apparent bigger infants (>1500 g 31 weeks’ gestation: 2.7 [2.5–3.0] vs 3.1 [2.4–3.8]; P = .007, adjustment weight age). CONCLUSIONS: Integration into programs may lead improved access secondary prevention facilitate assessment disease epidemiology resources.