Event Related Potential (P300) Detection using Convolutional Neural Network

Abstract

P300 event-related potential (ERP) signals are the cornerstone of the most important brain-computer interface systems. Those systems have become the most important means of communication between patients with neurological disorders and their external environment. This is due to the fact that P300 event potential (ERP) signals are an essential indication of studying and monitoring vital indicators of many diseases such as amyotrophic lateral sclerosis, Parkinson's disease, strokes, etc. Additionally, the P300 Event Potential (ERP) signals distinguish with their very stable latency with normal controls. In this study, the performance of one of the most important deep learning methods, convolutional neural networks (CNN), was evaluated. The P300 signal was classified through it in both the within-subject and cross-subject training conditions. An Electroencephalography (EEG) dataset consisting of 8 subjects was used. Windsorizing and a butter worse band pass filter were used as preprocessing techniques. Following filtering, the signals were fed into a convolutional neural network model for P300 classification and feature extraction. In the within-subject training scenario, the experimental findings showed that the convolutional neural network model was superior, achieving an average accuracy of 99.56%. Likewise, the average accuracy reached 89.45% in the case of cross-subject training.