Data integration aims to provide users with a unified, integrated and global view of the data stored in diverse, heterogeneous and autonomous sources. In the medical domain it is common to have information scattered in different sources such as medical images medical images, diagnostic reports and medical records. The integration of this information is important to support processes of diagnosis, treatment planning and monitoring, education and research. In this paper an integration architecture based on mediation is presented. The mediator uses a graph model to represent the structure of data contained in the sources. This is a conceptual model of the application domain and is formed by a set of subgraphs. Each subgraph represents the data of each data source. The sources have a metadata repository associated with data annotations, which are also used to access the data.