Decoding Multi-Class Motor Imagery From Unilateral Limbs Using EEG Signals.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Electroencephalography (EEG) is crucial for motor imagery-based brain-computer interfaces (MI-BCI), especially in stroke rehabilitation.
• Existing MI-BCI research primarily focuses on bilateral limb paradigms, posing challenges for unilateral upper limb applications.
• Decoding multiple unilateral motor imagery tasks is difficult due to overlapping neural activity.

Purpose of the Study:

• To develop a novel MI-BCI experimental paradigm for decoding multitasks in unilateral limbs.
• To evaluate the effectiveness of common machine learning techniques for this novel paradigm.
• To propose an advanced method for improving decoding accuracy in unilateral MI-BCI.

Main Methods:

• An experimental paradigm with four imagined unilateral limb movement directions was designed.
• Machine learning models including FBCSP, EEGNet, deepConvNet, and FBCNet were utilized for decoding.
• A novel MVCA method incorporating temporal convolution and attention mechanisms was proposed to enhance feature extraction.

Main Results:

• The MVCA model achieved classification accuracies of 40.6% for a four-class scenario and 64.89% for a two-class scenario.
• Decoding of specific diagonal movements (top-right to bottom-left and top-left to bottom-right) yielded the highest accuracies.
• This study is the first to demonstrate the successful decoding of motor imagery for multiple directions in unilateral limbs.

Conclusions:

• This research presents the first evidence for decoding motor imagery of multiple directions in unilateral limbs via EEG.
• The findings suggest that decoding diagonal movements offers the best accuracy, providing valuable insights for future research.
• This work advances MI-BCI paradigm development and demonstrates the feasibility of decoding complex directional information, expanding MI control command capabilities.