Exploring Markers of Brain-Computer Interface Performance in Children
Date
2024-05-14
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
For children with severe motor impairments, brain-computer interfaces (BCIs) are a potential life changing solution that provide an alternate means of communication and control. Some BCI users experience difficulties controlling a motor-imagery BCI (MI-BCI), but the unique factors influencing BCI performance in children are largely understudied. This study aimed to build predictive models of BCI performance in typically developing children using EEG correlates, demographic factors, and subjective assessments. We also aimed to explore specific features most predictive of BCI performance in children. Two datasets, DS1 (n=31) and DS2 (n=22) from independent studies comprising of EEG data, demographic information, and subjective assessments from typically-developing children were utilized. Models were trained 15 times, each on different feature subsets from DS1 using Support Vector Machine (SVM) and Random Forest (RF) classifiers, with hyperparameter optimization conducted using a differential genetic particle swarm optimizer. Models were internally (within-distribution) and externally (out-of-distribution using DS2) tested using repeated stratified k-fold cross-validation. We assessed the most commonly selected features for each feature set and the SHapley Additive exPlanations (SHAP) scores of the highest performing models. SVM models trained on EEG features outperformed models trained with all features or non-EEG features. The highest performing model was an SVM model trained on EEG features, with a cross-validation score, internal test score, and external test score of 0.76±0.24, 0.75, and 0.77 respectively. Connectivity and power in the alpha and theta bands were most predictive of BCI performance in our highest performing models. Our investigations provide a tool to predict BCI performance in children, and allow a fuller understanding of the internal factors influencing BCI performance. Predicting BCI performance in children can help to understand when standard BCI paradigms may not work for users and indicate when alternative methods of training may be necessary, ultimately improving the usability of BCIs in children.
Description
Keywords
Brain-Computer Interface, Pediatrics, Machine Learning, Electroencephalography, Motor Imagery
Citation
Kim, V. S. (2024). Exploring markers of brain-computer interface performance in children (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.