An Empirical Model-Based Algorithm for Removing Motion-Caused Artifacts in Motor Imagery EEG Data for Classification
Rajesh Kannan Megalingam1, Kariparambil Sudheesh Sankardas1, Sakthiprasad Kuttankulangara Manoharan1
1Humanitarian Technology (HuT) Labs, Department of Electronics and Communication Engineering, Amrita Vishwa Vidyapeetham, Amritapuri 690525, India.
Sensors (Basel, Switzerland)
|December 17, 2024
Summary
This study introduces a novel method to remove motion artifacts from electroencephalography (EEG) data for brain-computer interfaces (BCI). The new approach achieves 94.04% accuracy in classifying motor imagery (MI) EEG signals, aiding wheelchair users.
Area of Science:
- Biomedical Engineering
- Neuroscience
- Signal Processing
Background:
- Electroencephalography (EEG) is a valuable non-invasive tool for brain-computer interface (BCI) systems, particularly for individuals with severe mobility impairments.
- Motor imagery EEG (MI-EEG) data classification is crucial for BCI applications, enabling control of devices like wheelchairs.
- Motion artifacts significantly degrade EEG signal quality, posing a challenge for BCI systems used by mobile individuals, such as wheelchair users.
Purpose of the Study:
- To develop and validate an empirical error model-based artifact removal approach for cross-subject classification of MI-EEG data.
- To enhance the accuracy of MI-EEG classification for practical BCI applications, specifically for wheelchair users.
- To improve the decoding efficiency of motor imagery BCIs by addressing motion-induced artifacts.
Main Methods:
- Proposed an empirical error model incorporating inertial sensor data, wheelchair, subject weight, and terrain friction.
- Developed a modified Convolutional Neural Network (CNN)-based deep learning algorithm for MI-EEG classification.
- Recorded artifact data using three wheelchairs across five different terrains (road, brick, concrete, carpet, marble).
Main Results:
- Achieved a classification accuracy of 94.04% for distinguishing between four motor imagery classes (left, right, front, back).
- Demonstrated the effectiveness of the proposed CNN and empirical model in removing motion artifacts.
- Showcased superior performance compared to existing state-of-the-art techniques.
Conclusions:
- The proposed empirical error model-based artifact removal approach significantly improves MI-EEG classification accuracy.
- This method offers a potentially effective solution for enhancing BCI decoding efficiency for wheelchair users.
- The research facilitates practical BCI applications by enabling reliable interpretation of motor imagery signals despite motion artifacts.
