Deep Learning Models for Automated Diagnosis of Retinopathy of Prematurity in Preterm Infants

• DOI: 10.3390/electronics9091444
• Publication Date: 2020-09-04T15:24:24Z
• Authors: Yo-Ping Huang; Spandana Vadloori; Hung-Chi Chu; Eugene Yu-Chuan Kang; Wei-Chi Wu; Shunji Kusaka; Yoko Fukushima
• Keywords: Childhood blindness

Retinopathy of prematurity (ROP) is a disease that can cause blindness in premature infants. It characterized by immature vascular growth the retinal blood vessels. However, early detection and treatment ROP significantly improve visual acuity high-risk patients. Thus, diagnosis crucial preventing impairment. several patients refrain from owing to lack medical expertise diagnosing disease; this especially problematic considering number cases on rise. To end, we applied transfer learning five deep neural network architectures for identifying preterm Our results showed VGG19 model outperformed other models determining whether infant has ROP, with 96% accuracy, 96.6% sensitivity, 95.2% specificity. We also classified severity 98.82% accuracy predicting sensitivity specificity 100% 98.41%, respectively. performed 5-fold cross-validation datasets validate reliability found exhibited high ROP. These findings could help promote development computer-aided diagnosis.