MRCPs-and-ERS/D-Oscillations-Driven Deep Learning Models for Decoding Unimanual and Bimanual Movements

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Motor brain-computer interfaces (BCIs) aim to restore or compensate for central nervous system functions.
• Motor execution (ME) paradigm in BCIs utilizes residual or intact patient movement functions for intuitive control.
• Decoding unimanual and bimanual hand movements from electroencephalography (EEG) is crucial for assistance and neurorehabilitation, but multi-class classification remains challenging.

Purpose of the Study:

• To develop an improved deep learning model for decoding multi-class unimanual and bimanual hand movements from EEG signals.
• To enhance the performance of motor execution-based BCIs by integrating neurophysiological signatures.

Main Methods:

• Proposed a novel neurophysiological signatures-driven deep learning model.
• The model utilizes movement-related cortical potentials (MRCPs) and event-related synchronization/desynchronization (ERS/D) oscillations.
• The architecture includes a feature representation module, an attention-based channel-weighting module, and a shallow convolutional neural network.

Main Results:

• The proposed model achieved superior performance compared to baseline methods.
• Six-class classification accuracy for unimanual and bimanual movements reached 80.3%.
• Each feature module within the model demonstrated a significant contribution to the overall performance.

Conclusions:

• This study is the first to fuse MRCPs and ERS/D oscillations within a deep learning framework for enhanced motor execution-based BCI.
• The developed model significantly improves the decoding performance of multi-class unimanual and bimanual movements.
• This advancement can facilitate neural decoding for neurorehabilitation and assistive technologies.