MVAR ANALYSIS OF IEEG SIGNALS TO DIFFERENTIATE CONSCIOUS STATES

Neuroscience is a highly multidisciplinary and rapidly evolving research field. An important recent challenge of this discipline is the investigation of the so-called connectome. According to its original meaning, connectome is the map of the all brain neural connections. In this framework, the cognitive processes are not seen as localized in specific loci, but stored and processed in a distributed manner. Connectome aims to map and under-stand the organization of neural interactions trying, at the same time, to explain the role of functional units within the brain system. In particular, one of the most difficult and un-solved tasks in neuroscience is the identification of the areas, connections or brain func-tions that are called neuronal correlates of consciousness (NCCs). In this thesis the neural activity was explored by analysing human brain signals ac-quired during medical procedure. Signals from patients with drug resistant epilepsy were acquired by means of electrodes placed deep in the cortex (intracranial electroencephalog-raphy, EEG-iEEG), positioned in order to localize the epileptogenic focus. The technique, called stereotactic EEG (SEEG), guided and flanked by detailed 3D images, also pro-vides for periodical intracranial single-pulse electrical stimulation (SPES) to highlight are-as of interest. The continuous recording of the EEG activity took place for several days, and signals were grouped in two datasets: one acquired during wakefulness (WAKE) and the other one during the Non-Rapid Eye Movement sleep (NREM), stage 3. The signals were processed by means of two methods based on a multivariate auto-regressive model (MVAR). The first method was DTF (Directed Transfer Function), that is an estimator of the information flow between structures, depending on the signal fre-quency; it is able to describe which structure influences another. The second one was ADTF (Adaptive DTF) that permits to study the time-variant signal features, capturing their temporal dynamics. In addition to these connectivity analysis, feature extraction and classification techniques have been employed. The main aim of the dissertation is to evaluate methods and carry out analyses useful to distinguish between conscious and unconscious states, corresponding to WAKE and NREM respectively, studying at the same time the brain connectivity in response to Single Pulse Electrical Stimulation in intracranial EEG data. Massimini’s group (Department of Biomedical and Clinical sciences “L. Sacco”, Uni-versità degli Studi di Milano) revealed a different behavior for signals from the two states, WAKE and NREM: they noted a reactivation of the signal around 300 ms after the system perturbation in WAKE and, in contrast, a period of neural silence (down-state) in NREM condition. A hypothesis about the origin of the reactivation phenomenon is a feedback activity, i.e. the result of the activity from the rest of the network. In the thesis, the ADTF method was chosen to shed light on the down-state effect, paying attention to a defined temporal slice of data. The analysis was completed by the application of the DTF procedure, that was chosen to compare the two consciousness states and underline their differences in the frame of network connectivity. The analysis carried out lead to the following results:  Indication of useful combinations of features and techniques able to distinguish the states of interest  Observations of neural connection changes over frequency and time consider-ing causal relationships  Comparison of connectivity results using different re-referencing styles  Endorsement of the anatomical-functional importance of some channels corre-sponding to specialized brain areas. As conclusion of the analysis it was possible to identify a series of anatomical-functional brain features useful to discriminate the two mentioned states, therefore to speculate on the possibility to differentiate conscious and unconscious states with computational tools.