DL-Assisted ROP Screening Technique

• Published in: Biomedical Engineering Systems and Technologies Vol. 1710; pp. 236 - 258
• Main Authors: Kumar, Vijay, Patel, Het, Azad, Shorya, Paul, Kolin, Surve, Abhidnya, Chawla, Rohan
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2023; Springer International Publishing
• Series: Communications in Computer and Information Science
• Subjects: Computer Aided Diagnosis (CAD); Deep Learning (DL); Fundus image; Image processing; Retinopathy of Prematurity (ROP); YOLO-v5
• ISBN: 9783031206634; 3031206630
• ISSN: 1865-0929; 1865-0937
• DOI: 10.1007/978-3-031-20664-1_13

Retinopathy of Prematurity (ROP) is the most common cause of visual impairment among premature babies throughout the world. The consequence of ROP impairment can be minimized by performing suitable screening and treatment. However, due to deficiency of health care resources, many of these premature infants remain unidentified after birth. As a result, ROP-induced visual impairment is much more prevalent in these babies. We propose a robust and intelligent approach based on deep artificial intelligence and computer vision to automatically recognise the optical disk (OD) and retinal blood vessels and categorise the severe severity (Zone-1) of ROP patients in this study. We report empirical evidence using premature infant retina images from a nearby hospital to evaluate and validate the proposed approach. The YOLO-V5 prediction model identifies the OD from premature infants retina images, according to our results. Furthermore, the preterm infants’ fundus images were perfectly segmented by the computer vision-based system, which effectively separated the retinal vessels. Our system is able to obtain an accuracy of 82.5% in the Zone-1 occurrence of ROP.