Abstract: Federated learning is a bridge between medical data and machine learning, and can train data in a way that protects privacy. Personalized federated learning has better performance on the basis of its structure, especially when there are great differences in data distribution among different users. It not only realizes data privacy protection during training, but also achieves more superior performance, which has broad prospects in tasks such as intelligent diagnosis and auxiliary decision-making of multi center scenes. The development of personalized federated learning in medicine is discussed in this paper.