TSANet: multibranch attention deep neural network for classifying tactile selective attention in brain-computer...

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCIs) utilize electroencephalography (EEG) for noninvasive neural signal acquisition.
• Tactile-based BCIs offer an alternative to visual-based BCIs to mitigate visual fatigue.
• Current tactile-based BCIs exhibit unsatisfactory classification performance for control signals.

Purpose of the Study:

• To introduce a novel deep neural network, TSANet, for improved classification in tactile-based BCIs.
• To enhance the classification of tactile selective attention (TSA) using a multibranch convolutional neural network with a feature-attention mechanism.
• To evaluate TSANet's performance against conventional deep learning models.

Main Methods:

• Development of TSANet, a deep neural network incorporating multibranch convolutional neural networks and a feature-attention mechanism.
• Classification of tactile selective attention (TSA) signals within a tactile-based BCI system.
• Rigorous testing of TSANet under within-subject, leave-one-out, and cross-subject evaluation conditions.

Main Results:

• TSANet achieved superior classification performance compared to conventional deep neural network models across all evaluation conditions.
• The model demonstrated effective feature extraction for TSA by analyzing spatial filter weights.
• TSANet outperformed existing methods in classifying tactile selective attention signals.

Conclusions:

• TSANet presents a significant advancement for tactile-based BCIs, enhancing control signal classification.
• The proposed deep learning approach shows potential for efficient end-to-end learning in tactile BCI systems.
• Further development of TSANet could lead to more robust and user-friendly tactile BCIs.