Characterization of the spontaneous eeg activity in the alzheimer's disease continuumfrom local activation to network organization

Resumen

The human brain has billions of neurons with long and branching extensions that enable them to form connections and communicate with each other. The changes in these networks throughout the life span are responsible for physiological brain aging. However, normal brain aging can be altered by some physiopathological processes, like those associated with dementia due to Alzheimer's disease (AD) and mild cognitive impairment (MCI). AD is a progressive neurodegenerative disorder that causes cognitive, behavioral and functional alterations interfering with the patients' ability to carry out the activities of daily living. MCI is usually considered a prodromal stage of AD in which individuals show evidences of AD brain changes and exhibit a memory impairment beyond what would be expected for their age, but do not fully accomplish the criteria for dementia diagnosis. Despite the high prevalence of these neurodegenerative processes, the current knowledge about them is still limited. During the last decades, several neuroimaging techniques have been used to detect brain changes associated with neurodegeneration. Among them, electroencephalography (EEG) stands out for its high temporal resolution, necessary to study the dynamical processes involved in complex brain systems. Additionally, it is widely used in clinical settings due to its low cost and portability. In this context, the present Doctoral Thesis is focused on the characterization of the spontaneous EEG activity in the different stages of AD and MCI with the aim of gaining further insights into the neural substrates underlying their pathophysiology. For this purpose, two EEG databases have been studied, the former acquired in a clinical environment at the Río Hortega University Hospital and the latter acquired in a research environment as a result of a research project. Particularly, they have been investigated from three different levels of analysis depending on the way interactions are taken into account: (i) local activation; (ii) interaction between pair-wise electrodes; and (iii) network organization. Therefore, the contributions of the present Doctoral Thesis are canalized in these levels of analysis. In our first study, we developed and evaluated a three-way classification model with the aim of discriminating between AD patients, MCI subjects and healthy control (HC) subjects based on local activation features. Here, we take advantage of the complementary information provided by spectral measures and nonlinear methods to feed the classification model. These metrics also allowed us to investigate the changes in the EEG background activity in AD and MCI and study the abnormal brain changes associated with these neurodegenerative processes. Secondly, two studies that analyze the interactions between pair-wise electrodes were carried out. The first one was focused on the comparison of two coupling metrics derived from nonlinear methods. Only a few studies have applied entropy-based coupling metrics to biological systems, but they have never been used for the EEG characterization in AD. In this context, Cross-Approximate Entropy and Cross-Sample Entropy, and their parameter configurations, were compared to determine which of them yields more information about the abnormal coupling patterns in AD and MCI. In the second study, a new model to build synthetic signals based on the combination of a real-head surface-model and a set of coupled Kuramoto oscillators was proposed to study the behavior of different functional connectivity metrics under the effect of simulated volume conduction. Particularly, we evaluated how they are affected by volume conduction and if they are able to detect real changes in synchronization. Furthermore, the brain alterations in the different stages of AD continuum were characterized using the metric less affected by the influences of volume conduction (Phase Lag Index), thus reducing this bias. Lastly, we presented the design, development and evaluation of a new methodology to build multiplex network parameters based on the Canonical Correlation Analysis (CCA). CCA was used to determine the coefficients that maximize the correlation between electrode-level and source-level frequency-specific network parameters. In particular, we selected three parameters that are sufficiently representative and intuitive of the properties of the network and provide complementary information about how the brain network could work: strength, characteristic path length, and clustering coefficient. The behavior of the new CCA multiplex parameters was firstly studied using the synthetic signals generated with the Kuramoto model varying the level of connectivity. Then, the changes in the global brain network topology during the different stages of AD continuum were analyzed using these novel parameters. The application of these methods following the three levels of analysis led us to the main findings of the present Doctoral Thesis: (i) the local changes in the EEG background activity reflect a slowing of the EEG rhythms, a complexity loss, and a regularity increase, alterations that are related to the progression of AD; these changes were captured by a three-way classification model that showed a high diagnostic ability and culminated in a simplified clinical screening strategy for AD diagnosis (ii) a significant connectivity deficit in high frequency bands and a connectivity excess in low frequency bands were progressively found as the disease severity increases using entropy-based nonlinear metrics and Phase Lag Index; and (iii) the interpretation of the CCA multiplex parameters can be analogous to their frequency-specific counterparts and they revealed a loss of both integration and segregation in the multiplex brain networks as the disease progresses, making the AD networks more vulnerable as a consequence of the neurodegenerative process. These findings provide further and reliable insights into the brain alterations, useful to gain better understanding into the AD pathophysiology in its different stages. Furthermore, the novel models and methodologies proposed in the present Doctoral Thesis provide new perspectives for the study of the spontaneous EEG at the different levels of analysis.