Conclusions: Our supervised machine-learning methods that combine information from image and text are capable of achieving image annotation and retrieval accuracy acceptable for providing clinical evidence, given sufficient training data. (Source: International Journal of Medical Informatics).