Nowadays images are typically accompanied by additional information. At the same time, for example, magnetic resonance imaging exams typically contain more than one image modality: they show the same anatomy under different acquisition strategies revealing various pathophysiological information. The detection of disease, segmentation of anatomy and other classical analysis tasks, can benefit from a multimodal view to analysis that leverages shared information across the sources yet preserves unique information. It is without surprise that radiologists analyze data in this fashion, reviewing the exam as a whole. Yet, when aiming to automate analysis tasks, we still treat different image modalities in isolation and tend to ignore additional information. In this talk, I will present recent work in learning with deep neural networks, latent embeddings suitable for multimodal processing, and highlight opportunities and challenges in this area.
展开▼