A Few-Shot Transfer Learning Approach for Motion Intention Decoding from Electroencephalographic Signals.

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Decoding movement intention from electroencephalographic (EEG) signals is crucial for Brain-Computer Interface (BCI) development.
• Existing methods often require extensive training data for new tasks, limiting adaptability.
• Few-shot learning offers a promising avenue for reducing data requirements in BCI applications.

Purpose of the Study:

• To introduce and evaluate a few-shot transfer learning approach for decoding complex movement intentions from EEG signals.
• To develop a deep neural network (EEGframeNET5) capable of processing EEG signals in the space-frequency-time domain.
• To demonstrate the system's ability to adapt and recognize novel motor tasks with minimal training data.

Main Methods:

• A dataset of EEG signals for complex sub-movement preparation was curated.
• EEG signals were projected into the space-frequency-time domain and processed by a custom deep neural network (EEGframeNET5).
• A few-shot transfer learning strategy was employed to adapt the network for recognizing new, unseen tasks.

Main Results:

• The EEGframeNET5 achieved 72.45 ± 4.19% accuracy in classifying 5 classes from the source domain dataset.
• The few-shot transfer learning approach enabled the system to achieve 80 ± 0.12% accuracy in recognizing new tasks (hand opening/closing preparation).
• Performance in both phases surpassed comparable studies in the literature.

Conclusions:

• The proposed few-shot transfer learning methodology is effective for decoding motor preparation from EEG signals.
• This approach demonstrates significant potential for developing adaptive BCI systems for motion planning decoding.
• The methodology can be extended to other EEG-based applications like motor imagery and neural disorder classification.