Automated Early Detection of Retinopathy of Prematurity Zones using SWIN Transformer

Nazar Salih Abdulhussein; Royida  A. Ibrahem Alhayali; Mohammed Rashid  Subhi; Nebras  Hussein; Mohamed  Ksantini; Amina  Turki

doi:10.37934/ard.141.1.217231

Authors

Nazar Salih Abdulhussein Computer Science Department, Al-Imam Al-Adham University College, Baghdad, Iraq
Royida A. Ibrahem Alhayali Department of Computer Engineering, College of Engineering, University of Diyala, Diyala, Iraq
Mohammed Rashid Subhi Department of Petroleum System Control Engineering, College of Petroleum Processes Engineering, Tikrit University, Tikrit, Iraq
Nebras Hussein Biomedical Engineering Department, Al-Khwarizmi College of Engineering, University of Baghdad, Baghdad, Iraq
Mohamed Ksantini Control and Energies Management Laboratory (CEM-Lab), National Engineering School of Sfax, University of Sfax, Sfax, Tunisia
Amina Turki Control and Energies Management Laboratory (CEM-Lab), National Engineering School of Sfax, University of Sfax, Sfax, Tunisia

DOI:

https://doi.org/10.37934/ard.141.1.217231

Keywords:

Health care, deep learning, retinopathy of prematurity, fundus images, SWIN transformer

Abstract

Retinopathy of prematurity (ROP) is known to be the primary cause leading to permanent vision loss in children, which calls for its diagnosis and treatment based on subjective assessment of retinal vascular characteristics; even though this traditional approach is practical, it takes much time and likely results in errors. Therefore, automation is required not only to enhance precision but also productivity. The study proposes an innovative approach to early detection of ROP zones on fundus images between 2015 and 2020. It will use the Swin Transformer model, which has demonstrated superior precision and achieved a performance rate of 90.11%. This work denotes significant advancement in this field, emphasizing the potential of transformer-based architectures for the precise and efficient detection of ROP in clinical environments. The findings underscore the significance of utilizing state-of-the-art, comprehensive learning approaches to enhance early detection procedures, improving clinical outcomes for at-risk newborns.