Journal Home
Login
Register
Open Access Journal
Submit a Paper
Propose a Special lssue
Open Access
ARTICLE
Enhanced EfficientNet for Imbalanced Medical Image Classification through Grey Wolf Optimization
Issue Vol. 2 No. 02 (2025): Volume 02 Issue 02 --- Section Articles --- Published Date: 2025-08-17
Abstract
Medical image classification plays a pivotal role in modern diagnostics and disease management by aiding in the early detection and precise identification of various pathologies. However, a significant challenge in this domain arises from the inherent class imbalance commonly observed in medical datasets, where the number of samples for healthy cases or common conditions vastly outnumbers those for rare diseases. This imbalance often leads to deep learning models that are biased towards the majority class, resulting in suboptimal performance, particularly poor sensitivity and specificity for the critical minority classes. This article proposes a novel and robust approach to mitigate this pervasive issue by enhancing the state-of-the-art EfficientNet convolutional neural network (CNN) architecture through the application of Grey Wolf Optimization (GWO). GWO, a metaheuristic algorithm inspired by the sophisticated hunting strategies and social hierarchy of grey wolves in nature, is systematically employed to optimally tune the critical hyperparameters of the EfficientNet model. The primary aim of this optimization is to achieve superior and more balanced classification performance across all classes, especially for the underrepresented classes within imbalanced medical image data. We comprehensively detail the methodology, encompassing the meticulous handling and preprocessing of imbalanced medical image datasets, the strategic integration of the EfficientNet architecture, and the sophisticated GWO-based hyperparameter search strategy. Our experimental results, derived from rigorous evaluation, robustly demonstrate that the GWO-optimized EfficientNet significantly improves key performance metrics such as macro F1-score, balanced accuracy, and recall for minority classes. This optimized approach consistently outperforms traditional deep learning approaches that rely on manual hyperparameter tuning or fixed parameter sets. This research offers a robust, automated, and highly effective framework for developing more accurate, reliable, and clinically relevant deep learning models, thereby contributing significantly to the advancement of artificial intelligence in critical medical applications and enhancing diagnostic precision.
Keywords
References
[1] S. Liu, W. Wang, L. Deng, and H. Xu, “Cnn-trans model: a parallel dual-branch network for fundus image classification,” Biomedical Signal Processing and Control, vol. 96, Oct. 2024, doi: 10.1016/j.bspc.2024.106621.
[2] K. W. Goh et al., “Comparison of activation functions in convolutional neural network for poisson noisy image classification,” Emerging Science Journal, vol. 8, no. 2, pp. 592–602, Apr. 2024, doi: 10.28991/ESJ-2024-08-02-014.
[3] K. Man and J. Chahl, “A review of synthetic image data and its use in computer vision,” Journal of Imaging, vol. 8, no. 11, Nov. 2022, doi: 10.3390/jimaging8110310.
[4] E. T. A. Albert, N. H. Bille, and N. M. E. Leonard, “A mathematical primer to classical deep learning,” Journal of Applied and Advanced Research, vol. 9, pp. 15–25, Sep. 2024, doi: 10.21839/jaar.2024.v9.9169.
[5] A. Kaur and M. Kapoor, “An approach to recognize efficient deep learning model for pattern recognition,” in 2024 11th International Conference on Reliability, Infocom Technologies and Optimization (Trends and Future Directions) (ICRITO), Mar. 2024, pp. 1–6, doi: 10.1109/ICRITO61523.2024.10522108.
pdf
