Abstract
Medical image segmentation and classification have become fundamental components of modern intelligent healthcare systems due to their ability to support early disease diagnosis, treatment planning, prognosis evaluation, and computer-aided clinical decision-making. Recent advances in deep learning have significantly improved the capability of automated medical image analysis systems across multiple imaging modalities including X-ray, computed tomography, magnetic resonance imaging, ultrasound, histopathology, and retinal imaging. However, despite remarkable performance improvements, conventional deep learning models still face several limitations related to hyperparameter tuning, computational complexity, convergence instability, local minima stagnation, class imbalance, limited interpretability, and poor generalization under heterogeneous clinical conditions. To address these challenges, hybrid frameworks integrating deep learning with metaheuristic optimization techniques have emerged as a promising research direction for improving segmentation accuracy, classification robustness, feature optimization, and computational efficiency. This review article covers the historical evolution of intelligent medical image analysis, theoretical foundations of optimization-driven learning, taxonomy of deep learning architectures, and the role of evolutionary and swarm-based optimization algorithms including genetic algorithms, particle swarm optimization, grey wolf optimizer, firefly algorithm, whale optimization, ant colony optimization, Bayesian optimization, and reptile search optimization. Comparative analysis of datasets, evaluation metrics, computational complexity, convergence behavior, and clinical deployment challenges is also presented. Finally, open research challenges and future directions are identified toward trustworthy, interpretable, scalable, and autonomous AI-driven medical imaging systems for next-generation intelligent healthcare applications.

This publication has 54 references indexed in Scilit: