Decoding Changes in Tactile Stimuli Using Multivariate Pattern Analysis of EEG Data

Vibrotactile and multimodal displays are often designed to rely on the human ability to accurately detect when a change occurs within a display. This work investigates whether there is specific, objective information about the characteristics of tactile stimuli, that is, stimuli containing a change versus stimuli without a change, within event-related potential data from an EEG that can be extracted via multivariate decoding methods. A supervised machine learning classifier (i.e., support vector machine) was trained to classify change and no-change tactile stimuli based on the scalp voltage at electrode sites overlaying the somatosensory cortex. The classifier could predict with 63% accuracy whether or not a change trial had been presented to participants between 390 and 410 ms post-stimulus, demonstrating the feasibility of applying ERP- based multivariate decoding techniques to the study of tactile information processing. Therefore, the 390 to 410 ms post-stimulus time frame contains distinctive information that can be indicative of whether a change has occurred in tactile stimuli. The outcomes of this study can be broadly applied to advance brain-machine interfaces that seek to predict when specific tactile information is perceived and received, for example, for the development of adaptive displays in which information is selectively presented based on environmental or task conditions.