Predicting Retinopathy of Prematurity from Retinal Images of Preterm Infants Using Deep Learning Techniques

Abstract

Retinopathy of prematurity (ROP) is a proliferative vitreoretinopathy disorder newlinethat affects premature babies, which, when left untreated, can lead to permanent newlineblindness. It is characterised by retinal detachment in severe stages as a result of newlineaberrant cellular membrane development and contraction in the vitreous cavity as well newlineas on the sides of the retina. Due to developments in medical care for neonates, the newlinesurvival rates of preterm infants have increased tremendously. Due to high variability newlineand inter-observer inconsistency in the diagnosis and the increasing rate of ROP newlineincidence, there is a significant need to develop an automated system for the newlineprediction of ROP in the mass screening of infants. The earlier diagnosis of ROP is newlinebetter facilitated by applying Artificial Intelligence (AI) in its prediction from the noninvasive newlinefundus images of preterm newborns, which eases the stressful indirect newlineophthalmoscopic examinations by paediatric ophthalmologists. newlineThe first study aims to (i) develop a hybrid Deep Learning Network to predict newlineROP that can be used in the mass screening of infants and (ii) compare the newlineperformance of the proposed hybrid model with the Machine Learning (ML) newlineclassifiers and pre-trained Convolutional Neural Networks (CNN) model. The hybrid newlinenetwork is trained with 3200 and tested with 800 infant fundus images. Modified newlineMultiResUNet and matched filter with first-order Gaussian derivative are used to newlinesegment the retinal vessels from the fundus images. The Gray level co-occurrence newlinematrix (GLCM) and contour features of segmented images are extracted and selected newlineusing an embedded feature selection method. The selected features are evaluated newlineusing permutation importance and classified using the Random Forest (RF) classifier. newlineThe proposed hybrid model improves the quality of infant care by providing newlineadditional diagnostic assistance to clinicians. It also enhances the accessibility to newlineremote healthcare centres for large-scale automated screening systems newline