<p>Medical image segmentation plays a critical role in computer-assisted diagnosis, treatment planning, and clinical decision support systems. However, conventional deep learning-based segmentation models usually require large-scale annotated datasets that are not readily available in medical imaging due to the high cost of expert annotation and variability across imaging institutions. Meta-learning has been a promising paradigm for overcoming these limitations by enabling models to learn transferable knowledge across tasks and quickly adapt to new segmentation problems with limited training data. This review gives a comprehensive analysis of meta-learning approaches for medical image segmentation. The study provides a comprehensive review of existing research, in which meta-learning methods are classified into three main paradigms: metric-based, model-based, and optimization-based approaches. A comparative analysis is conducted of representative studies on multiple segmentation tasks, including organ, tumor, brain, prostate, and histopathological image segmentation, covering different imaging modalities such as magnetic resonance imaging (MRI), computed tomography (CT), and endoscopic imaging. The analysis reveals that optimization-based meta-learning methods exhibit high adaptability when combined with well-established segmentation architectures, whereas metric-based approaches are effective in few-shot scenarios with limited data. Despite these advances, there are challenges related to standardized evaluation protocols, cross-dataset generalization, and clinical deployment. The review identifies key methodological gaps and potential directions for future research to enhance the robustness, scalability, and real-world applicability of meta-learning-based medical image segmentation systems.</p>