This thesis presents an information fusion strategy for the automatic classification and retrieval of prostate histopathology whole-slide images (WSIs) that incorporates novel machine learning components from deep learning and kernel methods. Its main purpose is to enhance the representation of the WSIs using additional text content extracted from diagnosis reports. This is achieved using the multimodal latent semantic alignment (M-LSA) model, which employs a weakly-multimodal-supervised methodology that incorporates text information during the model training to enrich the representation of the WSIs with complementary semantic information. Besides, M-LSA does not require the text data during the prediction phase, which makes it suitable for realistic scenarios where a pathologist may only have the image data. The experimental evaluation demonstrates that the weakly-supervised multimodal enhancement has a significant improvement in the performance during classification and retrieval, further, the proposed model outperforms the state--of--the--art unimodal and multimodal baselines in automatic prostate cancer assessment.