Abstract: Medical image segmentation typically requires numerous dense annotations in the target domain to train models, which is time-consuming and labor-intensive. To mitigate this burden, unsupervised domain adaptation has been developed to train models with good generalisation performance on the target domain by leveraging a label-rich source domain and the unlabeled target domain data. In this paper, we introduce a novel Dynamic Prototype Contrastive Learning (DPCL) framework for cross-domain medical image segmentation with unlabeled target domains, which dynamically updates crossdomain global prototypes and excavates implicit discrepancy information in a contrastive manner. DPCL enhances the discriminative capability of the segmentation model while learning cross-domain global feature representations. In particular, DPCL introduces a novel crossdomain prototype evolution module through dynamic updating and evolutionary strategies. This module generates evolved cross-domain prototypes, facilitating the progressive transformation from the source domain to the target domain and acquiring global cross-domain guidance knowledge. Moreover, a cross-domain embedding contrastive module is devised to establish contrastive relationships in the embedding space. This captures both homogeneous and heterogeneous information within the same category and among different categories, enhancing the discriminative capability of the segmentation model. Experimental results demonstrate that the proposed DPCL is effective and outperforms the state-of-the-art methods.