Source-free domain adaptation for SSVEP-based brain-computer interfaces.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Brain-computer interfaces (BCI) using steady-state visually evoked potentials (SSVEP) aid individuals with speech impairments.
• Current SSVEP-BCI spellers often require lengthy calibration, causing user discomfort and hindering adoption.

Purpose of the Study:

• To develop a novel deep neural network (DNN) adaptation method for SSVEP-BCI spellers that eliminates the need for user-specific calibration.
• To enhance user comfort and accelerate the adoption of BCI technology.

Main Methods:

• Proposed a self-supervised deep learning approach to adapt a pre-trained DNN to a new user (target domain) using only unlabeled data.
• Introduced a custom loss function comprising self-adaptation (pseudo-labeling) and local-regularity terms to leverage data structure.

Main Results:

• Achieved high information transfer rates (ITRs) of 201.15 bits/min on the benchmark dataset and 145.02 bits/min on the BETA dataset.
• Demonstrated superior performance compared to existing state-of-the-art methods.

Conclusions:

• The proposed method significantly improves user comfort by removing the calibration requirement.
• This approach maintains high character identification accuracy and ITR, paving the way for wider BCI adoption in daily life.