A Robust 3D-Convolutional Neural Network-Based Electroencephalogram Decoding Model for the Intra-Individual...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Convolutional Neural Networks (CNNs) show promise for decoding electroencephalogram (EEG) in event-related potential-based brain-computer interfaces (ERP-BCIs).
• Traditional models struggle with long-term ERP-BCI performance due to intra-individual EEG variability over time.

Purpose of the Study:

• To propose and evaluate a novel three-dimension CNN (3D-CNN) model for dynamic EEG decoding.
• To address the limitations of static models in capturing time-varying EEG features for improved ERP-BCI robustness.

Main Methods:

• EEG data transformed into a time-series of brain topographic maps.
• A 3D-CNN model employed three-dimensional kernels to capture spatio-temporal EEG characteristics.
• Comparative analysis against 2D-CNN, LSTM, BP, and FLDA models using data from ten subjects over 6-12 hour sessions.

Main Results:

• The 3D-CNN model achieved higher decoding accuracies compared to baseline methods.
• The 3D-CNN exhibited a shorter computational cost than traditional models.
• Demonstrated advanced robustness against intra-individual EEG differences, crucial for long-term BCI applications.

Conclusions:

• The proposed 3D-CNN model offers a more practical and robust solution for long-term EEG decoding in ERP-BCIs.
• Dynamic decoding using 3D-CNN effectively handles intra-individual EEG variations.
• This approach enhances the reliability and performance of brain-computer interfaces.