Classification of Event Related Potential Patterns using Deep Learning

Abstract

Cognitive state of a person can be monitored by the use of brain electrical activity measurements (Electroencephalogram, EEG). In the concept of this study, it is aimed to classify EEG topographies using deep learning. Among the cognitive test paradigms, Stroop test with four colors is used to collect EEG from two participants. P300 and N400 components are selected as two classes. P300 topography is computed using the average of EEG from 280 to 320 ms after the stimuli while 380 to 420 time window is used for N400 topographies. After the EEG artefact rejection processes, 440 topograph images were used to train the deep network. Randomly selected 10 images that were excluded from training set were used for testing. All of the test images were correctly classified while 73% of the training set images were correctly classified.