Abstract: Pre-trained models have demonstrated broad application prospects in the medical field,with transfer learning and self-supervised learning providing novel technical paradigms for their pre-training process.However,unified evaluation standards and effective regulatory mechanisms are still lacking for medical pretrained models.This paper selects three representative tasks-medical visual question answering,medical report generation,and surgical workflow analysis-to review the application and evaluation of pretrained models in these tasks.On this basis,the evaluation metrics and test sets of existing models are further summarized,the major issues in current evaluation are analyzed,and the development trends are discussed,with the aim of promoting the refinement of evaluation methodologies for medical pre-trained models.