S12: Information fusion in Deep Learning for Biomedicine
Miguel Atencia
Francisco Veredas
Universidad de Málaga, Spain
Ruxandra Stoean
Romanian Institute of Science and Technology, Romania
University of Craiova, Romania
Abstract
The story goes that a picture is worth a thousand words and the success of Deep Learning techniques in image classification bears witness to the adage. Yet in many biomedical applications the availability of labeled images of diagnostic quality is rather limited as compared to, say, cats and dogs pictures. It comes as no surprise, then, that medical experts use a variety of media to diagnose, inform, and record conditions: images (magnetic resonance, tomography, ultrasound), signals (electrocardiogram, electroencephalogram), multi-omics data, and quite remarkably, natural language.
This special session aims at further exploring the application of Machine Learning methods to gain biomedical knowledge, emphasizing the integration of sources of information and highlighting the exploitation of data in scenarios where experience is scarce so far. The usage of word embeddings in natural language processing, echocardiography, fetal ultrasound, and analysis and prediction of COVID-19 data come to mind as examples, of course with no claim to be exhaustive. The scope of involved techniques is wide, including convolutional networks, statistical approaches, multi-layer perceptrons, and recurrent models.