Brain connectivity predicts performance in different domains of cognitive function in preschoolers

KAMIENKOWSKI, JUAN; PRATS, LUCIA; PIETTO, MARCOS L.; FRAIMAN, DANIEL; FRACCHIA, CAROLINA; HERMIDA, MARÍA JULIA; SIGMAN, MARIANO; LIPINA, SEBASTIAN J.

Buenos Aires

Congreso; 2nd FALAN Congress; 2016

FALAN

Brain networks have been extensively explored mostly in adults but also during development. These networks were defined on the basis of several measures of similarity between brain regions? activity from fMRI, MEG and EEG. In adults, it has been shown how different networks are related with different cognitive functions for instance the fronto-parietal network with top-down control. Thesefunctional networks initially found during the performance of a related cognitive task, also appears to be still coherent at rest. During development, several lines of research explore when and how these networks arose [2]. Interestingly, these studies typically explored linear correlations with age (but see [2]), while many plastic processes during first years of life depends nonlinearly on the age. Few studies focus on resting state EEG functional connectivity [3-5]. A major difference between this technique and fMRI is the opportunity to focus on different time scales or frequency bands, and thustarget different brain processes. They found significant differences in several frequency bands from 5 to 11 years using graph theory measures such as the characteristic path length, weight dispersion, and the modularity and homogeneity of the network [4,5]. These results also supported the generalpicture of segregation and integration drawn from the fMRI studies [1,2].1 Laboratorio de Inteligencia Artificial Aplicada, Depto. de Computación, Facultad de Ciencias Exactas y Naturales, Universidad de Buenos Aires,Argentina; 2 Unidad de Neurobiología Aplicada, CEMIC-CONICET, Argentina; 3 Departamento de Matemática y Ciencias, Universidad de San Andrés,Argentina; 4 Universidad Torcuato Di Tella, Argentina; 5 CONICET, Argentina.juank@dc.uba.ar Data acquisition: EEG activity was recorded @1024Hz over 128electrode positions on a standard 10?20 montage, using the BiosemiActive-Two system (Biosemi, Amsterdam, Netherlands). Preprocessing:The data was filtered between [0.2 100] Hz (notch: [49 51]Hz. Bad channels were detected by visual inspection and interpolated with the other channels weighted by the inverse distance. Artifactual ICs were selected and removed. We transform voltage to Current Source Density (CSD) [11] to subserve the independence of thesources. weighted Symbolic Mutual Information (wSMI) [12] is used as ameasure of connectivity between pairs of electrodes. We applied a lowpass filter at 10.7Hz, as suggested for tau=16 (and kernel=3). This filterincludes the low frequency bands (delta and theta) and the alpha peak,which is located around 8Hz for this age. Linear and Quadratic models: The association between the strength of a given link between two channels and a certain cognitive measure we evaluated the linear or quadratic coefficient resulting from the linear regression of cognitive measure and the wSMI. 95%CI were estimated for b1 and b2. We introduced an additional criteria for the quadratic terms to prevent that it only represented a small correction to the linear model and ensure that it was an U or inverse-U shaped relation.Permutation tests: The significance of the overall results were determined comparing with 1000 surrogates of the data. These surrogates were the result of the procedure explained before on 1000 permutations of the cognitive measures.Total number of significant connections (p