An Auxiliary Synthesis Framework for Enhancing EEG-Based Classification With Limited Data.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Deep learning significantly enhances brain-computer interface (BCI) performance using electroencephalogram (EEG) signals.
• High-resolution EEG data is crucial for training deep learning models, but collection is challenging due to subject burden and cost.

Purpose of the Study:

• To develop a novel auxiliary synthesis framework to overcome EEG data insufficiency for BCI.
• To generate artificial EEG data that preserves real data features and enhances model classification performance.

Main Methods:

• A framework combining a pre-trained auxiliary decoding model and a generative model was introduced.
• The framework learns latent feature distributions from real data and synthesizes artificial data using Gaussian noise.
• The method was evaluated on the BCI competition IV 2a dataset and compared against common data augmentation techniques.

Main Results:

• The proposed method effectively preserves time-frequency-spatial features of real EEG data.
• The synthesis framework significantly enhances classification performance with limited training data.
• The average accuracy of the decoding model improved by (4.72±0.98)% on the BCI competition IV 2a dataset, outperforming existing methods.

Conclusions:

• The novel framework offers an effective solution for generating artificial EEG signals to improve BCI classification performance.
• This approach reduces data acquisition costs and burdens in the BCI field.
• The framework is adaptable to various deep learning-based decoders and other BCI applications.