An inter-subject model to reduce the calibration time for motion imagination-based brain-computer interface.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Long calibration times for brain-computer interfaces (BCI) hinder practical application.
• Current BCI systems require extensive training data, leading to lengthy calibration stages for users.
• Individual variations in electroencephalogram (EEG) signals pose a challenge for cross-subject model training.

Purpose of the Study:

• To develop a new framework for reducing BCI calibration time.
• To improve the efficiency of BCI training by leveraging knowledge transfer from multiple subjects' EEG data.
• To address individual variations in EEG signals for more robust cross-subject model development.

Main Methods:

• A novel framework was proposed for BCI calibration time reduction using knowledge transfer from other subjects' EEG signals.
• Two data mapping techniques were introduced to mitigate variations in brain activation regions and strengths across subjects.
• An ensemble method was employed to aggregate processed EEG signals from multiple subjects into a unified model.

Main Results:

• The proposed method achieved satisfactory recognition accuracy with a minimal number of training trials (32 samples).
• The framework significantly outperformed existing methods that utilize few training trials.
• Data mapping techniques effectively reduced inter-subject variability, enhancing model generalization.

Conclusions:

• The proposed knowledge transfer framework effectively reduces BCI calibration time.
• The method demonstrates high recognition accuracy even with limited training data, making BCIs more accessible.
• This approach offers a promising solution for overcoming the calibration bottleneck in practical BCI applications.