<p>Lung disease detection has advanced rapidly with the adoption of artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL). However, traditional models often require large labeled datasets and struggle with generalization in real-world clinical settings. Meta-learning, or “learning to learn,”offers a promising solution by enabling models to adapt quickly with minimal data. This review presents a focused survey of meta-learning techniques applied to lung disease detection, including tuberculosis, pneumonia, COVID-19, and lung cancer. We examined 22 peer-reviewed studies to highlight methodological trends, commonly used datasets, and learning strategies. Few-shot learning methods, such as Prototypical and Siamese Networks, achieve an accuracy of over 85 % with as few as five samples per class, significantly outperforming standard DL approaches in low-data scenarios. Moreover, meta-learning can reduce the retraining needs by up to 60%, thereby enhancing the computational efficiency. We introduce a novel taxonomy tailored to medical imaging that addresses the limitations of traditional classifications (model-, metric-, and optimization-based). Our taxonomy organizes methods based on clinical context, data modality, task structure, and adaptability. This survey also identified open challenges such as computational cost, lack of interpretability, and limited domain generalization. Overall, meta-learning has emerged as a highly adaptable and efficient framework to improve AI-driven lung disease detection.</p>