Walking Imagery Evaluation in Brain Computer Interfaces via a Multi-View Multi-Level Deep Polynomial Network.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Rehabilitation Technology

Background:

• Brain-computer interfaces (BCIs) primarily focus on upper limb motor imagery (MI) due to challenges in detecting lower limb MI brain activity.
• Restoring or improving walking ability in disabled individuals requires reliable detection of lower limb motor control signals.

Purpose of the Study:

• To propose and validate a novel virtual environment (VE)-based walking imagery (WI) paradigm for eliciting reliable brain activity related to lower limb movement.
• To enhance the detection of brain activity for improved rehabilitation outcomes in individuals with lower limb impairments.

Main Methods:

• Extraction and fusion of spatial and time-frequency features to create a multi-view representation of brain activity patterns.
• Development of a multi-view multi-level deep polynomial network (MMDPN) to leverage feature complementarity for improved walking state detection.
• Comparison of the VE-based paradigm against traditional text-based paradigms using electroencephalography (EEG) data.

Main Results:

• The VE-based WI paradigm demonstrated significantly superior performance compared to traditional text-based paradigms.
• The VE-based approach effectively assisted users in modulating brain activity and improved the quality of EEG signals.
• The proposed MMDPN achieved superior classification performance over other deep learning methods.

Conclusions:

• The virtual environment-based walking imagery paradigm offers a promising approach for reliable lower limb motor control detection via BCIs.
• This novel paradigm has the potential to significantly advance the rehabilitation of walking ability for individuals with lower limb disabilities.
• The MMDPN provides an effective deep learning framework for analyzing complex brain activity patterns in BCIs.