2026-05-14T07:35:04.197Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:11439 2025-06-13T14:26:16Z 3

Ocular Disease Diagnosis Using CNNs Optimized by Modified Variable Neighborhood Search Algorithm, Chapter in AIS Algorithms for Intelligent Systems: IJCACI 2024: International Joint Conference on Advances in Computational Intelligence, Springer 2025 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/11439 https://link.springer.com/chapter/10.1007/978-981-96-3762-1_8 M. Zivkovic M. Antonijevic L. Jovanovic M. Krasic N. Bacanin T. Zivkovic M. Milovanovic Quick diagnosis of ocular diseases is important as numerous other ocular diseases, like diabetic retinophaty and glaucoma, that all may lead to irreversible vision loss if they are not treated early and properly. Therefore, quick diagnosis and early treatment can either slow or entirely stop the progression of the disease, preserving vision and patient’s quality of life. Moreover, early treatment is in most cases much more effective if the disease is caught in its early stages, like glaucoma. Finally, early and accurate diagnostics can aid in avoiding more complex and expensive treatment that may be necessary in the advanced stages of the ocular diseases, therefore effectively reducing the overall costs of healthcare. This study examines the utilization of convolutional neural networks (CNNs) to allow rapid and precise analysis of complex eye images of ocular disease patients, and help doctors with fast diagnostics. The CNNs are famous for processing immense volumes of images, however, they are heavily dependent on the hyperparameter’s values selection. This study proposes an altered version of variable neighborhood search algorithm (VNS) to tune the CNN’s hyperparameters for analysis of large datasets consisting of retinal images. The effectiveness of proposed approach is assessed in a real-world ODIR repository with very promising results. During executed experiments, the proposed approach is compared to other cutting-edge metaheuristics, and it outperforms all contending algorithms in this particular scenario, showing considerable potential in this domain. bookPart Springer, Singapore 99 112 10.1007/978-981-96-3762-1_8 AIS Algorithms for Intelligent Systems: IJCACI 2024: Proceedings of International Joint Conference on Advances in Computational Intelligence