Online adaptive classification system for brain–computer interface based on error-related potentials and neurofeedback. (April 2023)

• Record Type: Journal Article
• Authors: Haotian, Xu; Anmin, Gong; Jiangong, Luo; Fan, Wang; Peng, Ding; Yunfa, Fu
• Is Part Of: Biomedical signal processing and control. Volume 82(2023)
• Publication Date: 2023-04
• Subjects: Error-related potentials (ErrPs) -- Adaptive -- Neurofeedback (NF) -- Brain-computer interface (BCI) -- Channel-weighted common spatial pattern (CWCSP); Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2022.104554
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Before using the BCI system, a large amount of training data collection is usually required, which consumes much time and causes trial fatigue to subjects, but keeps the classifier constant after the training samples are collected. If the classifier needs to be updated, a new algorithm must be designed, or new training data must be collected. We proposes an online neurofeedback (NF) closed-loop system to verify the utility of ErrPs in a visual-motor imagery-based BCI system that continuously optimizes the classifier through the detection of ErrPs and improves the classification accuracy of the system. decoding results were corrected by detecting the ErrP signal of the subjects, the experimental data were self-verified with the corrected labels, and data that met the data quality standard were added to the training set, eliminating the need for extensive data collection before experiments, and enabling continuous updating of the training set as experiments proceeded. Furthermore, stable optimization effects were achieved in data analysis. We analyzed the accuracy enhancement of the adaptive classification system and investigated the neural mechanism behind ErrPs. Our proposed channel-weighted common spatial pattern (CWCSP) novel algorithm is combined with the ErrPs detection mechanism to realize a system that can be used with small training samples and optimize the classifier while automatically expanding the training samples as the experiment proceeds. Our study may provide new ideas for future smarter BCI systems, which not only can make the BCI system easier to use by eliminating the collection process of the training set, but also can update the decoding system of BCI for better performance according to the changes of the subject's brain signal state. Abstract: The electroencephalogram (EEG)-based brain–computer interface (EEG-BCI) is used in many fields, and can provide a more convenient way of life for patients with or without movement disorders. However, BCI systems are cumbersome in the training phase, and command recognition is subject to error. Error-related potentials (ErrPs) have been incorporated in BCI systems to enhance system performance. To this end, this paper proposes an online neurofeedback (NF) closed-loop system to verify the utility of ErrPs in a visual-motor imagery-based BCI system that continuously optimizes the classifier through the detection of ErrPs and improves the classification accuracy of the system. Our proposed channel-weighted common spatial pattern (CWCSP) novel algorithm is combined with the ErrPs detection mechanism to realize a system that can be used with small training samples and optimize the classifier while automatically expanding the training samples as the experiment proceeds. We analyzed the accuracy enhancement of the adaptive classification system and investigated the neural mechanism behind ErrPs. ErrP detection is used to correct decoding results so that the experimental data and the corrected labels are incorporated into the training set, and is then used to expand the training samples to optimize the classifier, showing good adaptive performance. This indicates that the system can effectively expand training samples and improve the decoding ability of a BCI system with a small number of training samples as the experiment progresses.