2026-05-04T19:52:44.561Z http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai

ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai:3:11665 2025-10-28T15:21:37Z 3

Ocular Disease Diagnosis Using Convolutional Neural Networks Optimized by Genetic Particle Swarm Optimization, Chapter in LNNS Lecture Notes in Networks and Systems: ICITI 2024: International Conference on Information Technology and Intelligence, Springer, volume 1341 2025 http://ezaposleni.singidunum.ac.rs/rest/sciNaucniRezultati/oai/record/3/11665 https://link.springer.com/chapter/10.1007/978-981-96-5126-9_4 S. Petrovic L. Jovanovic M. Antonijevic M. Milovanovic A. Jokic M. Zivkovic N. Bacanin M. Sarac Ocular diseases are common conditions that may severely affect a person’s quality of life. Rapid and accurate diagnostics are crucial for timely treatment. Utilization of well-trained machine learning models capable of fast data processing can reduce the time required for diagnosis significantly. Moreover, these models can attain accuracy levels that are comparable to human medical experts, which can further reduce the probability of a wrong diagnosis. This research explores the application of convolutional neural networks (CNNs) to accelerate the diagnostic process. However, this necessitates the selection of an adequate collection of hyperparameter values to obtain satisfactory results. Therefore, this study suggests the application of a modified form of notable particle swarm optimization (PSO) metaheuristics, that incorporates genetic operators, to select the appropriate values of CNN’s hyperparameters for this intrinsic task. Since the utilized dataset is imbalanced, the Matthews correlation coefficient (MCC) was employed as the objective function in the optimization process. Rigid comparative analysis between proposed PSO variation and other cutting-edge metaheuristics was shown, while the introduced model attained superior outcomes on the real-world ocular diseases dataset, achieving an accuracy of more than 61%, suggesting considerable potential in this domain. bookPart Springer, Singapore 43 57 10.1007/978-981-96-5126-9_4 LNNS Lecture Notes in Networks and Systems: ICITI 2024: Proceedings of International Conference on Information Technology and Intelligence, volume 1341