Computationally Enhanced Brain-Computer Interfaces Trained via Inverse Source Data

A brain-computer interface (BCI) is a communication system aimed to control a computer by means of neural activity. An application example of these systems is helping people with motor disabilities to drive a neurocognitive prostheses, just imagining the movements. These systems are mainly composed by: - a bench of sensors to monitor the brain activity; - a computational system to elaborate the obtained data and define the cognitive task performed by the subject; - a control system that exploit the elaborated data. The aim of this work is to improve the accuracy of the task classification in the computational system. The focus is in using EEG source imaging techniques to extract features to feed a machine learning classifier. Specifically, starting form a lead-field matrix, an algebraic model that map brain activity in the brain volume to the its surface, with an inversion of such matrix the volume activity is estimated from superficial EEG data. This work has investigated the impact on performance of the use of more accurate brain models and of related inversion strategies.