EEG-based AI-BCI wheelchair advancement: Transformer-based learning with motor imagery for brain computer interface

Area of Science:

• Neuroscience
• Artificial Intelligence
• Biomedical Engineering

Background:

• Brain-computer interfaces (BCIs) offer potential for assistive technologies.
• Controlling wheelchairs via BCIs requires accurate interpretation of neural signals.
• Motor imagery (MI) tasks are commonly used for BCI control.

Purpose of the Study:

• To develop an AI-integrated BCI system for wheelchair control using motor imagery.
• To classify electroencephalogram (EEG) signals corresponding to right- and left-hand movements.
• To evaluate the performance of a novel deep learning architecture, TFormerEEG.

Main Methods:

• Utilized a pre-filtered EEG dataset from an open-source repository.
• Segmented EEG data into 19x200 arrays sampled at 200 Hz.
• Developed TFormerEEG, a Transformer-based deep learning model for EEG classification.
• Integrated a Tkinter-based interface for wheelchair movement simulation.

Main Results:

• TFormerEEG achieved a test accuracy of 93.04% in classifying motor imagery EEG.
• The model outperformed baseline machine learning models like XGBoost and EEGNet.
• Stratified cross-validation yielded a mean accuracy of 91.18% for TFormerEEG.

Conclusions:

• The proposed TFormerEEG model demonstrates high efficacy for motor imagery EEG classification.
• The AI-integrated BCI system provides a functional and intuitive approach to wheelchair control.
• This approach holds promise for enhancing mobility and independence for individuals with disabilities.