Building a Tensor Framework for the Analysis and Classification of Steady-State Visual Evoked Potentials in Children

Steady-state visual evoked potentials (SSVEP) are one of several underlying signals used in various electroencephalography (EEG) based applications, including brain-computer interface (BCI) technology. Through oscillating visual stimulus at distinct frequencies, an SSVEP can be detected by EEG at occipital electrodes on the scalp, with distinct visual stimuli representing distinct choices. Rapid, accurate detection and classification of these signals is crucial for real-time analysis in SSVEP-based applications. However, signal analysis and interpretation of SSVEP events may be hindered in children due to the significant variability in electrophysiological signals throughout development. Recently, multi-way tensors have been shown capable of exploiting higher-order interactions present in the naturally multi-dimensional EEG data. Using tensors as tools to identify latent structures between varying maturational signals thus may provide a potential solution for rapid classification of SSVEP signals in children at different developmental stages. The presented methodology builds upon previous tensor-based SSVEP analysis and extends it for the first time to developing paediatric populations. Results from a binary SSVEP classification task of n = 40 children age 8-11 are reported to be significantly greater than chance, at 67-74% accuracy across multiple training and testing blocks. The findings support that tensor decomposition could provide flexible advantages capable of accommodating developmental differences across children and lay groundwork for future tensor analysis in SSVEP-based applications, like BCIs.