Statistical causality in the EEG for the study of cognitive functions in healthy and pathological brains

Understanding brain functions requires not only information about the spatial localization of neural activity, but also about the dynamic functional links between the involved groups of neurons, which do not work in an isolated way, but rather interact together through ingoing and outgoing connections. The work carried on during the three years of PhD course returns a methodological framework for the estimation of the causal brain connectivity and its validation on simulated and real datasets (EEG and pseudo-EEG) at scalp and source level. Important open issues like the selection of the best algorithms for the source reconstruction and for time-varying estimates were addressed. Moreover, after the application of such approaches on real datasets recorded from healthy subjects and post-stroke patients, we extracted neurophysiological indices describing in a stable and reliable way the properties of the brain circuits underlying different cognitive states in humans (attention, memory). More in detail: I defined and implemented a toolbox (SEED-G toolbox) able to provide a useful validation instrument addressed to researchers who conduct their activity in the field of brain connectivity estimation. It may have strong implication, especially in methodological advancements. It allows to test the ability of different estimators in increasingly less ideal conditions: low number of available samples and trials, high inter-trial variability (very realistic situations when patients are involved in protocols) or, again, time varying connectivity patterns to be estimate (where stationary hypothesis in wide sense failed). A first simulation study demonstrated the robustness and the accuracy of the PDC with respect to the inter-trials variability under a large range of conditions usually encountered in practice. The simulations carried on the time-varying algorithms allowed to highlight the performance of the existing methodologies in different conditions of signals amount and number of available trials. Moreover, the adaptation of the Kalman based algorithm (GLKF) I implemented, with the introduction of the preliminary estimation of the initial conditions for the algorithm, lead to significantly better performance. Another simulation study allowed to identify a tool combining source localization approaches and brain connectivity estimation able to provide accurate and reliable estimates as less as possible affected to the presence of spurious links due to the head volume conduction. The developed and tested methodologies were successfully applied on three real datasets. The first one was recorded from a group of healthy subjects performing an attention task that allowed to describe the brain circuit at scalp and source level related with three important attention functions: alerting, orienting and executive control. The second EEG dataset come from a group of healthy subjects performing a memory task. Also in this case, the approaches under investigation allowed to identify synthetic connectivity-based descriptors able to characterize the three main memory phases (encoding, storage and retrieval). For the last analysis I recorded EEG data from a group of stroke patients performing the same memory task before and after one month of cognitive rehabilitation. The promising results of this preliminary study showed the possibility to follow the changes observed at behavioural level by means of the introduced neurophysiological indices.