Tensor-driven extraction of developmental features from varying paediatric EEG datasets.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Data Science

Background:

• Childhood brain development presents unique challenges for electroencephalogram (EEG) technologies.
• Advanced signal processing is crucial for adapting EEG tools to pediatric populations.
• Tensor analysis offers a novel framework for analyzing complex, multi-dimensional EEG data.

Purpose of the Study:

• To explore tensor analysis for extracting developmental features from pediatric EEG datasets.
• To demonstrate the utility of identifying latent developmental characteristics in resting-state EEG.
• To improve the functionality and usability of EEG-dependent technologies for children.

Main Methods:

• Utilized a two-step constrained parallel factor (PARAFAC) tensor decomposition on three pediatric EEG datasets.
• Employed subject age as a developmental proxy.
• Applied Support Vector Machines (SVM) for age prediction and t-distributed stochastic neighbour embedding (t-SNE) for feature visualization.

Main Results:

• Successfully identified developmental features across different pediatric conditions.
• Achieved significant above-chance SVM classification accuracy in predicting subject age.
• Confirmed the critical role of tensor factorization in extracting relevant developmental EEG features.

Conclusions:

• Tensor analysis is a promising method for uncovering latent developmental EEG features in children.
• This approach can enhance the precision and applicability of EEG technologies in pediatric neuroscience.
• The findings support the use of advanced mathematical frameworks for understanding neurodevelopmental changes.