Temporal Convolutional Learning: A New Sequence-based Structure to Promote the Performance of Convolutional Neural Networks in Recognizing P300 Signals

Document Type : Original Article


1 Department of Computer Engineering, Faculty of Engineering, Alzahra University, Tehran, Iran

2 Department of Biomedical Engineering, Iranian Research Organization for Science and Technology, Tehran, Iran


Distinguishing P300 signals from other components of the EEG is one of the most
challenging issues in Brain Computer Interface (BCI) applications, and machine learning
methods have vastly been utilized as effective tools to perform such separation. Although
in recent years deep neural networks have significantly improved the quality of the above
detection, the significant similarity between P300 and other components of EEG in parallel
with their unrepeatable nature have led to P300 detection, which are still an open problem
in BCI domain. In this study, a novel architecture is proposed in order to detect P300 signal
among EEG, in which the temporal learning concept is engaged as a new substructure
inside the main Convolutional Neural Network (CNN). The above Temporal Convolutional
Network (TCN) may better address the problem of P300 detection, thanks to its potential
in involving time sequence properties in modelling of these signals. The performance of
the proposed method is evaluated on the EPFL BCI dataset, and the obtained results are
compared in two inter-subject and intra-subject scenarios with the results of classical CNN
in which temporal properties of input are not considered. Increased True Positive Rate of
the proposed method (an average of 4 percent) and its accuracy (an average of 2.9 percent)
in parallel with the decrease in its False Positive Rate (averagely 3.1 percent) shows the
effectiveness of the TCN structure in promoting the detection procedure of P300 signals in
BCI applications