Two-Stage Intelligent Multi-Type Artifact Removal for Single-Channel EEG Settings: A GRU Autoencoder Based Approach
IEEE Transactions on Bio-Medical Engineering
|March 24, 2022
Summary
This study introduces an intelligent system to remove physiological artifacts from single-channel Electroencephalogram (EEG) signals. The method accurately cleans EEG data corrupted by mixed artifacts, enhancing signal quality for wearable devices.
Area of Science:
- Biomedical Engineering
- Signal Processing
- Neuroscience
Background:
- Wearable and portable Electroencephalogram (EEG) systems face significant interference from physiological artifacts.
- Limited recording resources in these systems exacerbate artifact contamination in EEG signals.
Purpose of the Study:
- To investigate an intelligent artifact removal system for single-channel EEG signals.
- To address challenges posed by mixed multi-type artifacts in EEG recordings.
Main Methods:
- Representing mixed artifacts using unchanged latent pattern features in EEG signals.
- Employing an adaptive artifact removal scheme in the encoded feature domain.
- Formulating artifact removal as a two-stage identification-removal minimization problem with an attention-based adaptive feature concentration mechanism.
Main Results:
- Achieved 98.52% artifact identification accuracy on an open real-world dataset.
- Demonstrated an average correlation coefficient of 0.73 for removing strong, mixed multi-type artifacts.
- The system effectively handles single-channel EEG signals with high accuracy and low overhead.
Conclusions:
- The developed system is effective and stable for removing mixed multi-type artifacts from single-channel EEG signals.
- It offers superior performance compared to traditional schemes with fixed criteria.
- This technology can significantly improve EEG signal quality for simplified sensing systems.
