Few-shot Medical Classification: Methods, Challenges and Future Directions

In the context of medical imaging, due to the high cost and time-consuming nature of expert annotation, the scarcity of available annotated data often leads to overfitting during network construction. Few-shot learning offers a promising solution to this issue and has attracted significant research attention in recent years. This survey provides a comprehensive review of the most advanced techniques in few-shot learning-based medical image classification. Depending on the underlying deep learning mechanisms, existing methods are categorized into four groups: transfer learning-based, meta-learning-based, data augmentation-based, and multi-modal-based methods. This article also systematically summarizes publicly available few-shot medical image datasets and outlines prospective research directions.