Sub-band classification of decomposed single event-related potential co-variants for multi-class brain–computer interface: a qualitative and quantitative approach

• Published in: IET science, measurement & technology Vol. 10; no. 4; pp. 355 - 363
• Main Authors: Ahirwal, Mitul Kumar, Kumar, Anil, Singh, Girish K, Suri, Jasjit S
• Format: Journal Article
• Language: English
• Published: The Institution of Engineering and Technology 01.07.2016
• Subjects: artificial neural network; Artificial neural networks; BCI system design; brain‐computer interfaces; Classification; clinical diagnoses; Decomposition; Electroencephalography; Enterprise resource planning; event‐related potential; Feature extraction; generalised EEG system; Human-computer interface; medical signal detection; multiclass brain‐computer interface; multiclass subband classification; neural nets; power spectrum‐based feature extraction; signal classification; single event‐related potential covariant decomposition; single‐trial ERP detection; Systems design; electroencephalography; artificial neural network; brain-computer interfaces; single event-related potential covariant decomposition; single-trial ERP detection; decomposition; multiclass subband classification; medical signal detection; event-related potential; clinical diagnoses; signal classification; feature extraction; multiclass brain-computer interface; power spectrum-based feature extraction; BCI system design; neural nets; generalised EEG system
• ISSN: 1751-8822; 1751-8830
• DOI: 10.1049/iet-smt.2015.0200

Subject dependent and co-variant nature of electroencephalography (EEG)/event-related potential (ERP) are still hurdles in the development of generalised EEG system such as clinical diagnoses and brain–computer interface (BCI) systems. Presently, classification of BCI classes is limited due to consideration of several trials of ERP as average case for generalisation or some time single-trial ERP detection is preferred for subject specific schemes. Present study tries to develop a methodology for utilisation of co-variant nature of ERP, based on the sub-band decomposed ERPs. Proposed idea also explores a new dimension in BCI system design, as multi-class classification for execution of multiple commands through single ERP. Sub-band ERPs and their power spectrum-based features are extracted and classified successfully with 70.64% accuracy using artificial neural network.