Classifying the auditory P300 using mobile EEG recordings without calibration phase.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Mobile Brain Computer Interfaces (BCI) typically require extensive subject-specific training data, hindering immediate user interaction.
• The calibration phase for supervised classification is a significant drawback in current BCI systems.

Purpose of the Study:

• To develop a subject-independent classification method for mobile EEG BCI.
• To enable immediate interaction for new users by removing the need for supervised classification and calibration.

Main Methods:

• Utilized canonical polyadic decomposition (CPD) to exploit structural differences in three-class auditory oddball data.
• Incorporated average event-related-potential (ERP) templates into the CPD model.
• Developed a novel similarity measure between single-trial pairs and known templates for classification.

Main Results:

• Achieved classification accuracy comparable to supervised, cross-validated stepwise Linear Discriminant Analysis (LDA).
• Demonstrated a method that does not require subject-dependent data.
• The CPD method provides a fast and interpretable classifier.

Conclusions:

• The proposed CPD method offers a significant practical advantage over traditional BCI approaches.
• Subject-independent BCI classification is feasible, enhancing user accessibility.
• This novel approach paves the way for more immediate and user-friendly BCI applications.