Meeting Banner
Abstract #3548

A Support Vector Machine Based Real-Time fMRI Communication Channel

Tom Ash1, Adrian Carpenter1, Guy Williams1

1Wolfson Brain Imaging Centre, University of Cambridge, Cambridge, United Kingdom

We present a support vector machine based technique for allowing subjects in a scanner to communicate yes / no answers to questions, with answers available to experimenters in real time (less than 1 TR). The technique uses SVMs to detect subject brain state as they perform mental imagery to reply to questions. Tests in healthy volunteers show 19% of subjects answers were not distinct enough to be confidently labeled. Of the remaining 109, 107 answers were correctly interpreted by the classifier. This technique shows promise for use as a real-time communication channel for behaviorally vegetative, internally conscious brain injury patients.