Using Hierarchical Models to Understand P300-Wave--Based Brain-Computer Interface Performance Among Disabled Adults

• Activity Number: 322
• Type: Contributed
• Date/Time: Tuesday, August 2, 2016 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Maryclare Griffin* and Peter Hoff
• Companies: and University of Washington
• Keywords: eeg ; hierarchical ; tensor ; array ; event-related potential ; classification

Abstract

Brain-computer interfaces, specifically P300 spellers, are designed with the goal of allowing subjects to communicate letter-by-letter via EEG waves. They work be leveraging a known physiological response, the P300 wave, which appears in EEG recordings when a rare stimulus is presented. As brain-computer interfaces have developed, many statistical methods for detecting the P300 response have appeared. Unfortunately, despite the fact that EEG data are often collected on several subjects and multiple EEG channels simultaneously, most existing statistical methods are trained on data from single subjects and aggregated EEG channels. Consequently, these methods fail to perform well on the disabled subjects for whom they would be most beneficial. To explore the cause of poor performance among disabled subjects, we develop a new statistical model for detecting P300 waves that shares information both across subjects and EEG channels.