Boosted diagnostic accuracy in glaucoma detection with SqueezeNet-TR lite architecture and precise optic disc localization
Alexandria Engineering Journal · 2025-11 · 131 · 48-89
DOI: 10.1016/j.aej.2025.10.007 ↗Kurz gefasst
Fachartikel in der Zeitschrift Alexandria Engineering Journal, 2025-11. Verfasst von Mohamed Mouhafid und 3 weiteren Autor:innen. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 54 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).
Abstract (Originalsprache)
Glaucoma, the leading cause of irreversible blindness, requires early and accurate detection for effective treatment planning. Existing automated systems encounter several limitations, including insufficient accuracy in detecting subtle glaucoma-related changes, computational inefficiency, reliance on full-fundus views, class imbalance, and lack of interpretability, hindering their clinical application. This study proposes an enhanced SqueezeNet architecture for glaucoma detection (GD) that incorporates grouped convolutions, simplified fire modules, and transformer encoders to improve feature extraction and long-range dependency modeling. A novel multi-scale adaptive template matching method is introduced for precise optic disc (OD) localization, eliminating irrelevant background information. Class imbalance is addressed using a hybrid SMOTE-TOMEK strategy to enhance the model sensitivity. Model interpretability is improved through saliency maps, promoting confidence in clinical application. The architecture is computationally efficient, with only 0.8 million parameters, ensuring practicality for large-scale deployment. It achieved an accuracy of 99.52 % on REFUGE, 99.15 % on ACRIMA, and 94.23 % on RIM-ONEv2, demonstrating robustness across diverse datasets. Compared with 13 well-known convolutional neural network (CNN) models, the enhanced architecture outperformed most of them across multiple metrics. With improved accuracy, efficiency, and OD-focused interpretability, this method enables practical, large-scale glaucoma screening for clinical and telemedicine use.
Zitieren
APA
Mouhafid, M., Zhou, Y., Shan, C., & Xiao, Z. (2025). Boosted diagnostic accuracy in glaucoma detection with SqueezeNet-TR lite architecture and precise optic disc localization. Alexandria Engineering Journal, 131, 48-89. https://doi.org/10.1016/j.aej.2025.10.007
Vancouver
Mouhafid M, Zhou Y, Shan C, Xiao Z. Boosted diagnostic accuracy in glaucoma detection with SqueezeNet-TR lite architecture and precise optic disc localization. Alexandria Engineering Journal. 2025;131:48-89. doi:10.1016/j.aej.2025.10.007
BibTeX
@article{Mouhafid2025,
author = {Mouhafid, Mohamed and Zhou, Yatong and Shan, Chunyan and Xiao, Zhitao},
title = {Boosted diagnostic accuracy in glaucoma detection with SqueezeNet-TR lite architecture and precise optic disc localization},
journal = {Alexandria Engineering Journal},
year = {2025},
volume = {131},
pages = {48--89},
doi = {10.1016/j.aej.2025.10.007}
}
EUVIMED