An improved P300 extraction using ICA-R for P300-BCI speller

• Main Authors: Lee, Wee Lih, Tan, Tele, Leung, Yee Hong
• Format: Conference Proceeding
• Language: English
• Subjects: Data mining; Electroencephalography; Independent component analysis; Integrated circuits; Testing; Training; Vectors
• ISSN: 1094-687X; 1558-4615; 2694-0604
• DOI: 10.1109/EMBC.2013.6611185

In this study, a new P300 extraction method is investigated by using a form of constrained independent component analysis (cICA) algorithm called one-unit ICA-with-reference (ICA-R) which extracts the P300 signal based on its temporal information. The main advantage of this method compared to the existing ICA-based method is that the desired P300 signal is extracted directly without requiring partial or full signal decomposition and any post-processing on the outcome of the ICA before the P300 signal can be obtained. Since only one IC is extracted, the method is computationally more efficient for real-time P300 BCI applications. In our study, when tested on the BCI competition 2003 dataset IIb, the current state-of-the-art performance is maintained by using the one-unit ICA-R. Besides that, the ability of the method to visualize P300 signals at the single-trial level also suggests it has potential applications in other types of ERP studies.