A Multi-Channel Ensemble Method for Error-Related Potential Classification Using 2D EEG Images

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Error-related potentials (ErrPs) are neural signals indicating a mismatch between expected and actual outcomes.
• Accurate ErrP detection is vital for enhancing the performance and user experience of brain-computer interfaces (BCIs).

Purpose of the Study:

• To propose and validate a novel multi-channel method for improved ErrP detection using a 2D convolutional neural network.
• To enhance the accuracy and reliability of ErrP classification in BCI systems.

Main Methods:

• A multi-channel approach integrating multiple classifiers for ErrP detection.
• Transformation of 1D electroencephalography (EEG) signals from the anterior cingulate cortex (ACC) into 2D waveform images.
• Classification using an attention-based convolutional neural network (AT-CNN) and a multi-channel ensemble approach.

Main Results:

• The proposed AT-CNNs-2D method achieved an accuracy of 86.46%, sensitivity of 72.46%, and specificity of 90.17%.
• The multi-channel ensemble approach demonstrated a 5.27% higher accuracy compared to majority voting ensembles.
• Validation on two datasets confirmed the effectiveness of the proposed method for ErrP classification.

Conclusions:

• The AT-CNNs-2D model significantly improves ErrP classification accuracy in BCI applications.
• The developed multi-channel ensemble method offers a robust approach for integrating channel-specific decisions.
• This research provides innovative strategies for advancing ErrP detection in brain-computer interfaces.