EEG Cortical Source Feature based Hand Kinematics Decoding using Residual CNN-LSTM Neural Network.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Signal Processing

Background:

• Motor kinematics decoding (MKD) from brain signals is crucial for developing brain-computer interface (BCI) systems for rehabilitation and prosthetics.
• Surface electroencephalogram (EEG) is commonly used for MKD, but decoding from cortical sources remains less explored.
• Utilizing pre-movement brain activity for MKD offers potential for more intuitive and responsive BCI control.

Purpose of the Study:

• To investigate the feasibility of decoding hand kinematics using EEG cortical source signals.
• To explore the utility of pre-movement EEG data for MKD.
• To evaluate a deep learning model for hand kinematics decoding from cortical sources.

Main Methods:

• A residual convolutional neural network (CNN) - long short-term memory (LSTM) model was developed.
• The model utilized pre-movement EEG cortical source signals, specifically windows 50 ms prior to movement onset.
• Hand kinematics decoding was performed for grasp and lift tasks, with performance assessed using correlation values (CV) in both sensor and source domains.

Main Results:

• The proposed deep learning model successfully decoded hand kinematics using pre-movement EEG cortical source data.
• The study demonstrated the viability of using pre-movement neural information for MKD.
• Performance comparison between sensor and source domains indicated the effectiveness of the source-based approach.

Conclusions:

• Hand kinematics can be effectively decoded from pre-movement EEG cortical source signals.
• This approach holds promise for enhancing the capabilities of BCI systems.
• Further research into cortical source analysis for BCI applications is warranted.