Subject-Transfer Decoding using the Convolutional Neural Network for Motor Imagery-based Brain-Computer Interface.

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Motor imagery (MI)-based brain-computer interfaces (BCIs) utilize pattern-recognition and machine learning to interpret brain signals.
• Current MI-BCI methods often require extensive training data, limiting real-world applicability.

Purpose of the Study:

• To develop a subject-transfer decoding method for MI-BCI that reduces training time and maintains accuracy.
• To enhance the robustness of MI-BCI systems with limited training data.

Main Methods:

• A convolutional neural network (CNN) was employed for subject-transfer decoding.
• The CNN was pre-trained on MI data from multiple subjects and then fine-tuned to a target subject's data.
• The BCI competition IV data2a, featuring 4-class MIs, was used for evaluation.

Main Results:

• The proposed CNN method demonstrated superior accuracy compared to the self-training method across varying numbers of training trials.
• Accuracy improvements ranged from 4.94% (288 trials) to 12.31% (144 trials) over the self-training approach.
• The method maintained classification accuracy effectively even with significantly reduced training trials.

Conclusions:

• The subject-transfer CNN decoding method is effective for MI-BCI applications.
• This approach significantly reduces the need for extensive subject-specific training data.
• The findings suggest a more practical and efficient implementation of MI-BCIs in real-world scenarios.