In the latest years, new deep learning models have been proposed in the literature, called Generative Adversarial Neural Networks (GANNs), that turned out as effective at synthesizing high-quality images in several domains. We study novel applications of GANNs to the automatic generation of medical images to augment biomedical datasets, translate images into other domain, reconstruct missing parts.

MRI data are usually represented as images. However, new data representation approaches were developed based on graph theory. Recently applied in neurosciences, graph-based models opened new perspectives for the exploration of brain structural and functional connectivity by means of graph-derived metrics.

Data-driven disease classification can be extremely useful in medical diagnosis. However, the massive amount of information provided by gene expression or clinical data might decrease the accuracy of the classifier. We study a combination of data reduction and data visualization techniques to facilitate the task of disease classification and provide an early diagnosis.

In medical imaging, semantic segmentation can recognize different regions that can correspond to: tissue classes, pathologies, organs, other biologically relevant structures, etc. It is a challenging task, due to noise, smoke, blur, specular reflections, occlusions.

Continual Learning (CL) aims at defining models that continuously learn and evolve according to new amounts of data, retaining previously learned knowledge. CL approaches can be of great help in Healthcare, as they can mitigate the catastrophic forgetting of old concepts when new ones are learned, and help at learning new tasks without using training data for previous tasks, which might no longer be accessible.