Advanced Computational Models for Neuroinformatics and Brain-Computer Interfaces

Neuroinformatics is an emerging interdisciplinary field aiming to develop methods, algorithms, and visualisation tools for the study of the brain and nervous system. This discipline intersects the field of brain-computer interfaces (BCIs), whose purpose is to exploit external signals causing detectable physiological reactions in the brain or to recognise signals produced by aware activities in the brain as an alternative communication output toward the external world. This thesis uses electroencephalography (EEG) to measure the activity of the human cerebral cortex related to specific brain functions: movements, sleep, and emotions. In addition, it presents advanced computational models to process, analyse, and decode the resulting signals, which are nonlinear and nonstationary and are greatly affected by noise and artifacts. The final goal is to decode the unknown signals of emotions, one of the most complex neurocognitive domains due to its extreme subjectivity and the impossibility of verifying the labels in the collected data. This is done first by validating the proposed models on data allowing well-known brain functions, such as movements and sleep, and then by applying them to the complex and subjective signals from emotions. An evaluation of the activation protocols used to elicit brain activity related to the above functions is furnished, and specific activation protocols are presented for collecting bias-reduced, specific datasets. Experimental results and comparisons with state-of-the-art models are presented and discussed, highlighting the pros and cons of the proposed strategies and suggesting some hints for future developments.