A review and systematic guide to counteracting medical data scarcity for AI applications

Artificial intelligence has the potential to improve the scalability, objectivity, and precision of the overall healthcare system. Such improvements are possible due to the growth of medical databases and the progress of deep learning approaches, which enable automated analysis of both structured and unstructured data. While the overall size of medical datasets continues to increase, data scarcity remains problematic due to challenges in the medical domain, such as rare diseases, difficult and expensive annotation, and restricted population coverage. Machine learning models trained without appropriate measures to counteract this scarcity are often biased and unreliable in real-world settings. This paper will systematically examine the different challenges arising from medical data scarcity, their implications, and state-of-the-art mitigation approaches. It includes studies from the general machine learning community and describes how their findings translate to medical applications. This review is meant as a practical resource for researchers who want to develop reliable machine learning models for medical applications when data is scarce. • Explores challenges of medical data scarcity and their real-world implications. • Comprehensive review of state-of-the-art data- and model-centric solutions. • Highlights solutions for imbalanced, limited, and noisy datasets. • Actionable insights for mitigating bias and improving reliability.