Discriminative Canonical Pattern Matching for Single-Trial Classification of ERP Components

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Event-related potentials (ERPs) are key brain-computer interface (BCI) control signals.
• ERPs are weak and sensitive to experimental variations, leading to diverse patterns across datasets.
• Developing general decoding algorithms for varied ERPs with small training sets remains challenging.

Purpose of the Study:

• To compare the performance of Discriminative Canonical Pattern Matching (DCPM) against seven other ERP-BCI classification methods.
• To evaluate the robustness of DCPM in classifying diverse ERP components using small training datasets.

Main Methods:

• Single-trial classification of ERPs from two private and three public EEG datasets.
• Comparison of DCPM with Linear Discriminant Analysis (LDA) variants, Spatial-Temporal Discriminant Analysis (STDA), xDAWN, and EEGNet.
• Utilized ERPs including P300, motion visual evoked potential (mVEP), and asymmetric visual evoked potential (aVEP).

Main Results:

• DCPM demonstrated superior classification performance across all tested datasets.
• The algorithm showed robustness in handling diverse ERP components and small training set limitations.
• DCPM outperformed LDA variants, STDA, xDAWN, and EEGNet.

Conclusions:

• DCPM is a robust and effective algorithm for classifying a wide range of event-related potentials in BCI applications.
• The findings suggest DCPM's potential for developing more adaptable and generalizable BCI decoding algorithms.
• This study highlights DCPM's advantage in scenarios with limited training data and diverse ERP characteristics.