Elucidating age-specific patterns from background electroencephalogram pediatric datasets via PARAFAC

Brain-computer interfaces (BCI) have the potential to provide non-muscular rehabilitation options for children. However, progressive changes in electrophysiology throughout development may pose a potential barrier in the translation of BCI rehabilitation schemes to children. Tensors and multiway analysis could provide tools which help characterize subtle developmental changes in electroencephalogram (EEG) profiles of children, thus supporting translation of BCI paradigms. Spatial, spectral and subject information of age-matched pediatric subjects in two EEG datasets were used to form 3-dimensional tensors for use in parallel factor analysis (PARAFAC) and direct projection comparison. Within dataset cross-validation results indicate PARAFAC can extract age-sensitive factors which accurately predict subject age in 90% of cases. Cross-dataset validation revealed extracted age-dependent factors correctly identified age in 3 of 4 test subjects. These findings demonstrate that tensor analysis can be applied to characterize the age-specific subtleties in EEG, which provide a means for tracking developmental changes in pediatric rehabilitation BCIs.