Role of spatial filtering in the pre-processing chain of a BCI for non-responsive patients

Brain injuries represent a relevant current public health problem. Often, they lead to periods of unconsciousness, coma, or even death. Generally, individuals who exit from a period of coma survive in a non-responsive state, unable to voluntarily respond to external stimuli. For these nonresponsive patients, even though they accomplish movements, it is not trivial to establish and associate them with a level of consciousness, in order to interpret their intentions. This obstacle may easily conduct to misleading diagnoses. In order to obtain a more reliable diagnosis, it is essential to establish and develop methods able to enhance studies of various levels of consciousness in nonresponsive patients. In this thesis an innovative brain-computer interface (BCI) is presented, having the goal to analyze electroencephalographic (EEG) and electromyographic (EMG) signals, recorded from healthy individuals, to extract meaningful information about the intention to move while performing or imagining to perform motor tasks and to compare them with the ones retrieved from nonresponsive patients. This BCI is characterized by a pre-processing chain that aims at cleaning the raw EEG signals, extracting particular voltage waves called readiness potential (RP), which are analyzed and improved to be sent to a machine-learning algorithm, which will utilize them to perform a classification to execute differential diagnosis. The thematic nucleus of this thesis regards spatial filtering, which represents one of the blocks of the pre-processing chain that consists of an offline method to filter data in the spatial domain. Most specifically, the main analyses will deal with: (1) a definition of the idea behind a spatial filter with a presentation of several pre-existent versions in literature; (2) a classification of their principal features; (3) an explanation of updates introduced in the previous versions of spatial filters to try improving them; (4) several discussions about their performance in extracting a motor component called Lateralized Readiness Potential (LRP), in generating topographies of the scalp and modifying RPs of healthy people subjected to voluntary, semi-voluntary and involuntary tasks; (5) final reflections regarding their role with advantages and disadvantages they bring to final RPs for machine learning algorithm. In addition to that, this BCI is employed to study signals registered from hemiplegic patients to verify whether the motor intention on the hemiplegic side is preserved. Therefore, spatial filters are tested on data recorded from hemiplegic, as well.