Automatic Identification of Brain Independent Components in Electroencephalography Data Collected while Standing in a...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Electroencephalography (EEG) is vital for brain activity monitoring, but artifact removal is time-consuming.
• Automating EEG processing is essential for real-time brain-computer interface (BCI) applications.
• Subject-specific variations in independent components (ICs) necessitate standardized artifact identification methods.

Purpose of the Study:

• To develop and evaluate a deep learning classifier, DeepIC-virtual, for automated identification of brain components versus artifacts in EEG data.
• To assess the feasibility of deep learning techniques for classifying ICs from noisy EEG data collected in a virtual reality (VR) environment.

Main Methods:

• A convolutional neural network (CNN) deep learning model was trained on 1432 manually labeled IC images (topographical maps) from EEG data.
• EEG data was collected from six subjects standing in a VR setup with simulated height/depth variations and perturbations.
• The CNN model performed binary classification to distinguish between brain components and artifactual ICs.

Main Results:

• The DeepIC-virtual model achieved a classification accuracy of 89.20% and an area under the curve of 0.93.
• The model demonstrated high performance even with imbalanced data, misclassifying only 1 out of 57 brain ICs in the test set.

Conclusions:

• DeepIC-virtual shows significant viability for automated classification of artifactual ICs in EEG data.
• This automated approach can facilitate the integration of BCI methods for clinical applications, such as anxiety control and acrophobia treatment.