Large collections of medical images have become a valuable source of knowledge, taking an important role in education, medical research and clinical decision making. An important unsolved issue that is actively investigated is the efficient and effective access to these repositories. This work addresses the problem of information retrieval in large collections of biomedical images, allowing to use sample images as alternative queries to the classic keywords. The proposed approach takes advantage of both modalities: text and visual information. The main drawback of the multimodal strategies is that the associated algorithms are memory and computation intensive. So, an important challenge addressed in this work is the design of scalable strategies, that can be applied efficiently and effectively in large medical image collections. The experimental evaluation shows that the proposed multimodal strategies are useful to improve the image retrieval performance, and are fully applicable to large image repositories.