Individualized electrode subset improves the calibration accuracy of an EEG P300-design brain-computer interface for people with severe cerebral palsy

Introduction This study examined the effect of individualized electroencephalogram (EEG) electrode location selection for non-invasive P300-design brain-computer interfaces (BCIs) in people with varying severity of cerebral palsy (CP) in a post-hoc offline analysis. Methods A forward selection algorithm was used to select the best performing eight electrodes (of an available 32) to construct an individualized electrode subset for each participant. Custom electrode subset size was chosen to be 8 because BCI accuracy of the individualized subset was compared to accuracy of a widely used default subset. Results Across 51 participants, individualized subsets improved calibration accuracy only for the severe CP cohort (mean +28.6% absolute; 95% CI 13.4%, 46.1%; p 0.0001). No group-level benefit was detected for mild CP or typically developing controls, although several individuals in these groups improved (2/17 mild CP; 1/10 controls). In the subset with held-out testing data (mild CP and controls), calibration gains did not translate to higher testing accuracy; among controls, the subset effect was reduced on testing (−9.6%, 95% CI −13.3%, −5.8%, p 0.0001), with no evidence of change for mild CP. Participants with severe CP typically required larger subsets to approach asymptotic accuracy, whereas ≤ 8 electrodes were sufficient for most others. Discussion The findings suggested that electrode selection can accommodate atypical neuroanatomy in people with severe CP, while the default electrode locations are sufficient for people with milder impairments from CP and typically developing individuals.