Yksilöllinen variaatio aivojen oskillatorisessa aktiivisuudessa nREM unen aikana
Heikkinen, Verna · Aaltodoc (Aalto University) · 2022
Sleep electroencephalogram (EEG) is known to be highly individual and heritable, with unique aspects related to maturation. The magnetoencephalographic (MEG) spectral power structure has recently been associated with genetic factors using Bayesian reduced-rank regression (BRRR) for extracting a low-dimensional representation of familial data features. This approach has shown that the spectral power structure is highly consistent within participants irrespective of experimental state, suggestive of subject-specific cortical fingerprints. How well the algorithm generalizes to more noise-prone and low-resolution clinical EEG data and, e.g., during brain maturation is not known. In this thesis BRRR was applied on 19-channel non-REM sleep EEG recordings of ~800 healthy Finnish children, between 3 weeks -- 19 years of age, for finding the spatiospectral components that would maximally differentiate subjects from each other. 2--4 spectral estimates per participant were used for addressing the accuracy of the model, by calculating L1 distances between participants in the low-dimensional latent space. The test subjects could be separated from each other with > 73\% average accuracy using 12 spatiospectral components. The components explaining the most variation correlated moderately with age, possibly reflecting individual trajectories in brain maturation. The latent space components were generally not consistent over participants’ own data sets from different sleep phases, but restricting analysis to the oldest subjects (>7 years) yielded within-subject accuracy of 65\%. Future studies are needed for revealing the possible benefits of using higher-density EEG recordings and/or of combined longitudinal M/EEG recordings in addressing cortical fingerprints in children. The promising results of this thesis encourage the use of BRRR for EEG data and also in clinical research, where additional confounders could be used to study brain function in different neurological disorders.