Decoding motion prediction using EEG pattern classification

Abstract

Due to the processing delay of the visual system, we would expect the perception of moving stimuli to lag behind their actual location. This is not the case: we are able to react to the veridical location of moving objects. Evidence suggests that motion prediction is possible due to extrapolation, but not much is known about the neural mechanisms underlying this phenomenon. We used multivariate pattern classification on EEG data to decode motion prediction. Pattern classification has recently gained much traction as a method for studying time sensitive neural mechanisms in the brain. We trained a classifier on easy-to-decode EEG data. Using this classifier, we compared time-to-peak latencies in classifier performance between an apparent motion condition, where motion extrapolation is possible, and a scrambled condition, where motion extrapolation is not possible. We found faster time-to-peak latencies for the apparent motion condition. This is in agreement with results found in previous experiments, and adds to the body of evidence for motion extrapolation as mechanism for motion prediction. We also take a short look at why linear discriminant analysis was used as our classifier.