Automatic EEG classification during rapid serial visual presentation task by a novel method based on dual-tree complex wavelet transform and Poincare plot indices

Abstract

Nowadays, brain-computer interfaces (BCI) are designed to do the desired action only by analyzing brain signals. Taking the advantages of computers, these systems have higher reliability and speed than conventional mankind approaches. In a typical BCI system, the stimulation is applied and at the same time, the participants’ brain signals are recorded. The system then performs the desired action by decoding the brain responses. One of the most important stimuli is a rapid serial visual presentation (RSVP). This study dealt with the problem of automatic classification of the target/non-target images with different presentation rates at RSVP events using electroencephalogram (EEG) data. In the present work, we propose a combined dual-tree complex wavelet transform (DTCWT) and Poincare plot indices to robustly characterize and classify EEG responses elicited during an RSVP experiment. Additionally, a robust classification scheme was developed by taking advantage of feature selection algorithms and a probabilistic neural network (PNN). The features were classified into six categories, including target/non-target events displayed at the rate of 5 Hz (C1/C2), target/non-target events displayed at the rate of 6 Hz (C3/C4), and target/non-target events displayed at the rate of 10 Hz (C5/C6) using one vs. all strategy. The framework was evaluated using EEG time-series from an RSVP task database available at Physionet. The data were collected from eight channels while target and non-target pictures were shown in different displaying rates, including 5, 6, and 10 Hz. The results showed the robust maximum average classification rates in the range of 81.5 to 83.33% for all feature selection strategies. In conclusion, the proposed framework paved the way for designing a reliable BCI system for target detection and classification.