From single layer to multilayer networks in mild cognitive impairment and Alzheimer’s disease

--

We investigate the alterations of functional networks of patients suffering from mild cognitive impairment and Alzheimer's disease (AD) when compared to healthy individuals. Departing from the magnetoencephalographic recordings of these three groups, we construct and analyse the corresponding single layer functional networks at different frequency bands, both at the sensors and the regions of interest (ROI) levels. Different network parameters show statistically significant differences, with global efficiency being the one having the most pronounced differences between groups. Next, we extend the analyses to the frequency-band multilayer networks (MN) of the same dataset. Using the mutual information as a metric to evaluate the coordination between brain regions, we construct the αβ MN and analyse their algebraic connectivity at baseline λ2−BSL (i.e., the second smallest eigenvalue of the corresponding Laplacian matrices). We report statistically significant differences at the sensor level, despite the fact that these differences are not clearly observed when networks are obtained at the ROIs level (i.e., after a source reconstruction procedure). Next, we modify the weights of the inter-links of the multilayer network to identify the value of the algebraic connectivity λ2−T leading to a transition where layers can be considered to be fully merged. However, differences between the values of λ2−T of the three groups are not statistically significant. Finally, we developed nested multinomial logistic regression models (MNR models), with the aim of predicting group labels with the parameters extracted from the MN (λ2−BSL and λ2−T). Using these models, we are able to quantify how age influences the risk of suffering AD and how the algebraic connectivity of frequency-based multilayer functional networks could be used as a biomarker of AD in clinical contexts.