Disorder of consciousness evaluation with functional connectivity indexes applied to EEG signals

Despite the many definitions through which consciousness can be described, from an anatomical point of view it can be seen as the integration of two main components: the state of Alertness and the Awareness content. Studies concerning consciousness can be better carried out analyzing the Alertness component, as its state results more detectable compared to awareness. Consciousness is characterized by the activation of the brain stem and the cerebral cortex due to the thalamus. Its main functions rely on the definition of the sleep-wake cycle and, at the same time, on different sensory and cognitive applications. The study of wakefulness allows to distinguish intermediate states between sleep and wake, and they might not always be physiological. In fact, should a subject undergo a severe brain injury, their consciousness might get impaired. The main goal of this work is to assess the Disorder Of Consciousness (DOC) of patients on different severity levels using statistical metrics and Functional Connectivity (FC) indexes applied to signals deriving from one of consciousness’ primary centers, that is the cerebral cortex. Comparing pathological results with healthy subjects is fundamental to classify and distinguish different severity levels on patients suffering from DOC. There is not just a single tool capable of studying the state of consciousness of a patient, but ElectroEncephaloGraphy (EEG) recordings, especially when compared to MRI, resulted to be the most suited for the case thanks to their high temporal resolution which allows to precisely localize a phenomenon in time. For this work, EEG was acquired from the patients’ scalp and the study was carried out only on the meaningful channels common to every subject. The signals were properly filtered to extract both EEG’s whole frequency range and the single frequency bands composing it. For a study of this kind, certain frequency bands, in fact, might be more informative than others, as they might be present only for either awake or unconscious subjects. Thus, every measure applied to the signals was evaluated for all the EEG frequency bands. Further pre-processing was needed to remove ElectroOculoGraphy (EOG) and ElectroMyoGraphy (EMG) artifacts. Approximate Entropy (ApEn) and Phase-Amplitude Coupling (PAC) were computed as the main statistical measures for each of the frequency ranges of the signals. Then, Graph theory was applied to recreate the brain network based on correlation coefficients obtained from all channel combinations. The assessed coefficients were: Mutual Information, Covariance, Pearson’s Correlation coefficient, Cross-correlation, Phase-Locking Value, Phase Lag Index. Adjacency matrices were so obtained, in both binarized and weighted versions, and FC indexes were consequently calculated. The most informative indexes resulted to be global indexes such as Characteristic Path Length and Global Efficiency, and local indexes like Clustering coefficient and Local Efficiency. Entropy, PAC and FC indexes showed different values when healthy subjects were compared to the whole group of pathological ones. In particular, the latter showed larger standard deviation values and more outliers. No clear differences appear, though, when trying to distinguish between severity levels of DOC, especially when comparing the lower levels. EEG is, in any case, an unpredictable tool affected by its random nature, but the collected data set the stage for the creation of an automatic classifier for patients with a DOC.