A study on Retinopathy of Prematurity(ROP) Screening Using Deep Learning Approaches and a Blood Vessel Segmentation Framework for ROP Affected Eyes

The retina contains millions of photoreceptive cells called cones, rods, and blood vessels that nourish these cells. Retinopathy of prematurity is an eye disease affecting the retina of a prematurely born infant and is one of the leading causes of childhood blindness at a global level. Pediatric fundus images captured using retina imaging devices such as Retcams have been used by doctors for analyzing the retina and detecting various abnormalities present in the retina. Multiple stages in a single ROP image, underdeveloped retina vessels, immaturity of the macula and fovea,image quality issues like illuminations and artifacts present in an image make ROP disease detection, classification, and segmentation a difficult task for both an experienced practitioner and a computer-aided diagnostic system(CAD). Nowadays automated image analysis using artificial intelligence techniques is gaining popularity among practitioners and researchers. The objective of this survey is to do a systematic review of published works that use deep learning approaches for disease detection, classification, and segmentation in the context of Retinopathy of prematurity (ROP). This work presents a performance analysis and comparison of state-of-the-art deep learning architectures us ed for the above tasks. We also provide potential avenues for future research in ROP detection, classification, and segmentation using deep learning. After the study we proposed a segmentation framework called ROP-seg using a publicly available dataset for blood vessel segmentation that uses the combined power of Unet and Efficientnet architectures. The proposed system shows promising results in blood vessel segmentation of ROP affected eyes.